Ataköy Görüntüleme: SEO and AI-Search Audit

1 Executive decision

Ataköy Görüntüleme has a substantial service catalogue, accessible HTML pages, a physical location and several ways to make contact. The public site also contains procedure mix-ups, conflicting equipment descriptions, incomplete language journeys and measurement that does not reliably distinguish a contact click from a completed request. I would address those issues before expanding acquisition.

The fresh crawl reached all 180 sitemap-listed URLs and six additional linked destinations. Every destination ultimately returned HTTP 200. This changes the technical diagnosis: the evidence supports a publishing, routing and measurement repair programme. It does not establish a site-wide indexing failure or a need to replace WordPress. The ranking gap with HSM Radyoloji has different causes, explained in the supplement at the end: HSM has a separate page for each examination and about ten times as many sites linking to it, while mixed-up copy on the clinic’s pages teaches search engines the wrong topics.

The first decisions

• Assign a clinical owner to correct wrong-procedure and conflicting equipment copy across Turkish, English and Arabic pages. Confirm actual service availability before promoting it.
• Repair conversion definitions and review consent handling. An appointment-link click emitted an Ads conversion request before any form submission. Consent and form-permission observations are covered in a separate internal annex.
• Fix shared navigation and complete one language-consistent service journey at a time. Several links return a working page with the wrong purpose.
• Choose the first service improvements using confirmed capacity, provider-selection demand and qualified enquiry data. Coronary CT, MRI, HSG, dental CT and breast ultrasound are candidates, not an approved revenue forecast.
• Close the ranking gap where it starts: give each priority examination its own accurate page, keep the Google Business Profile consistent with the site, and earn legitimate links from referrers, insurers and local organisations (supplement).
• Establish native search and AI visibility baselines. The 2025 promotion regulation largely rules out paid search and sponsored social campaigns aimed at patients in Türkiye, so organic search and the Google Business Profile are the main digital channels to invest in (section 17).

What a successful first release looks like

The clinic can point to medically approved service information, a correct contact route in the selected language, intentional consent behaviour and a verified request-success event. Management can then compare qualified enquiries and attended examinations by service without treating every click as a patient.

Independent public-site audit by Alex Rostovtsev, prepared before any engagement. Research dated 13–15 October 2026, with re-checks on 16 October. This report evaluates public communication and acquisition systems; it does not evaluate clinical care. Privacy, security and regulatory observations are in a separate internal annex, not in this report. AI and search answers from eight engines were collected on 17 October 2026.

Evidence E01–E11. Exact URL observations and implementation checks appear in the accompanying findings register.

2 How the audit layers fit together

I would keep these workstreams separate in the scope and connect them through one commercial objective: help a suitable person or referrer find accurate service information and make a request the clinic can handle. Technical SEO supports both conventional and AI search. Measurement and privacy run across every channel.

Priorities describe decisions

P1 means correct or verify before scaling the affected journey. P2 means the next planned improvement after the immediate gates. P3 means routine hygiene or a lower-confidence investigation. There is no invented overall health score: an accessible page with the wrong procedure and a missing description on a legal page should not carry equal weight.

Evidence labels

Observed means a saved public response, rendered page, network request or ad-library record supports the statement. Reported means the imported workbook or earlier research supplies the number. To verify means an account setting, operational fact or causal interpretation still requires evidence. Recommendations do not imply that an observed defect caused a particular ranking, lost patient or legal breach.

The report has a matching CSV register so a developer, editor or clinic owner can take responsibility for individual actions without working through the narrative each time.

3 Research scope and limits

What was inspected

• Public HTTP crawl: 180 sitemap URLs plus six eligible linked destinations; 196 sequential requests including robots, seven sitemap files and redirects. The 186 records contain 185 HTML responses and one KML response; 183 distinct final HTML pages.
• Rendered browser and network checks: Turkish, English and Arabic homepages; sampled MRI, booking and price-request pages; cookie states before choice, after rejection and after acceptance; one ordinary appointment-link click. Additional mobile booking inspection used a narrow-screen view; the saved DOM records a 375px document width with no horizontal overflow.
• Content and intent: fresh service/about/contact samples; all crawl source pages for titles, headings, canonical, language, structured data and internal links; six generic Turkish search samples across five priority families; two limited provider-page comparisons.
• AI readiness: a completed Blursor homepage check, with its request ledger and rendering limitation retained. Current official Google, Bing, OpenAI and Anthropic guidance was checked.
• Historical demand: both supplied workbooks and the reconstructed prior conversation. The corrected high-intent v2 workbook takes precedence over the initial broad-volume recommendations.
• Paid evidence: public ad-library records, reviewed in the internal annex; no private performance accounts.
• Re-checks on 16 October 2026: Ahrefs organic and link metrics for both domains, HSM’s top pages, Ataköy Görüntüleme’s ranking keywords and selected live pages.

What the evidence cannot establish

There was no authenticated Search Console, GA4, Google Ads, Business Profile, Bing Webmaster Tools, CRM, server-log or financial access. No patient data was entered, no patient form was submitted, no result portal was opened and no ad account was created. Actual indexing, verified crawler behaviour, current campaign status, cost per qualified enquiry and clinical capacity remain account or owner checks.

The search samples were generic web searches, not a location-configured Google rank tracker or Maps grid. No site-wide Core Web Vitals field dataset, repeated Lighthouse benchmark, complete accessibility assessment or security audit was performed. The mobile observations support specific form improvements, not a universal device or WCAG verdict.

Reproducibility

The evidence archive retains source URLs, timestamps, crawl CSV/JSON, selected HTML, scripts, request summaries and screenshots. Diagnostic requests identify the test infrastructure; successful requests using a crawler-like user agent do not authenticate a vendor crawler. Retest the important journeys after changes and before using this dated report as a current status statement.

E01 public crawl; E02 browser/network evidence; E03 imported demand research; E04 content/search samples; E05 Blursor; E06 official documentation.

4 Demand and commercial priorities

The imported Ahrefs work reports 19 total ranking keywords in the Türkiye top-100 report, with no verified non-brand split. Site Explorer estimated 59 organic visits per month; a separate Batch Analysis view estimated 34. These third-party snapshots do not measure actual traffic, local discovery, referrals or patient volume. They support investigating missed provider-selection demand without quantifying the clinic’s total marketing performance. A matched 15 October snapshot and the reasons behind the gap with HSM are in the supplement.

The corrected workbook tested 406 exact action and geographic combinations. It labels 70 as positive volume, 52 as explicit zero and 284 as no matching Ahrefs row. A blank numerical cell must not erase the difference between zero and unavailable data. The local tab substantially reuses the probe; its 272 rows cannot be added to the 406.

Reported Ahrefs estimates from 14 October 2026. These are Türkiye-wide exact-query estimates, including queries containing İstanbul. Variants overlap. Do not total them or convert them into local searches, clicks or patients.

Intent needs a second check

Fresh search samples mixed providers, directories, explainers and discussions. A phrase containing a booking verb can still include public-hospital or informational intent. The workbook’s price query sanal anjiyo fiyatı has a reported 1,900 searches, but interest in prices is not authorization to publish healthcare price promotions.

Choose the first two journeys with the clinic

Confirm examination availability, spare capacity, reporting staff, referral prerequisites, contribution after variable costs and the team’s ability to answer requests. Prioritize the intersection of service fit, credible demand and operational readiness. Keep open MRI and detailed obstetric claims out of targeting until verified. Add varicose-vein or broader laboratory campaigns only after the actual service owner and workflow are established.

E03 high-intent shortlist and probe; E04 six search samples. The first workbook remains useful as a broad catalogue, with the later intent correction preserved.

5 What priority service pages should answer

The current pages often explain a modality more clearly than they explain a visit to Ataköy Görüntüleme. The first content work should make verified operational information usable. Preserve useful existing URLs while deciding which page owns each service intent.

A reusable service brief

For each confirmed service, publish the location, actual service scope, named reporting responsibility where appropriate, referral/document requirements, approved preparation route, appointment process, realistic total visit and result windows, supported languages and how questions are handled. Explain how to obtain an accurate individual quote or insurance clarification only within the clinic’s legally reviewed publication rules.

Preparation, contrast, sedation, screening and suitability decisions belong to the clinical team. This audit does not supply patient instructions or assume that a scanner specification determines the appropriate examination.

Two comparison pages used for information design

Memorial’s coronary CT page demonstrates the usefulness of distinguishing preparation, scan duration, total visit time and report timing. Derya’s dental-imaging page distinguishes external patients, images, reports and delivery. These are examples of questions to answer with Ataköy Görüntüleme’s own facts, not evidence of those providers’ traffic, clinical quality or compliance.

Comparison sources: www.memorial.com.tr · view source and www.deryaklinik.com · view source. E04 retains the retrieval limits; Memorial text was available through web extraction while a separate direct GET returned 403.

6 Correct procedure and equipment copy first

F01 Published pages mix unrelated procedures

Observed, P1. The Turkish ultrasound page /ultrasonografi/ includes claustrophobia, sedative and open-MRI guidance in its ultrasound explanation. The passage was confirmed in a rendered browser view. /usg-ultrasonografi/ also contains an MRI examination list in its HTML source. The English fluoroscopy page /en/fluoroscopy-2/ introduces unrelated colonoscopy and MRI material and mistranslates imaging terminology.

The issue also affects /bt-cihazi/ and its Arabic counterpart /ar/bt-cihazi-ar/. Their title, main heading and explanation refer to different procedures; the body discusses abdominal ultrasound, names a different imaging centre (“Sonomed Tıbbi Görüntüleme Merkezi ve Laboratuvarı”) as the provider and includes MRI preparation. This proves inconsistent published information. It does not establish how the copy originated or assess care delivered by the clinic. The same claustrophobia and sedation paragraph also appears on /enteroklizis/. This is a ranking problem as well as a trust problem: on 16 October, Ahrefs showed the enteroclysis page ranking for “emar ilacı” (position 5) and “mr korkusu için ilaç” (position 8). Search engines are learning the wrong topic for these pages.

Action: the clinical lead and editor should identify the intended purpose of each page, verify service availability, replace unrelated blocks and review adjacent templates for the same reuse. Give patients one coherent explanation of the confirmed procedure.

Acceptance: each retained page describes one approved service consistently; preparation belongs to that service; attribution identifies the correct organisation; the clinical owner signs off the source and each translation. Do not redirect a page until its intended purpose and current search value are understood.

F02 Equipment descriptions contradict their labels

Observed, P1. /manyeti-rezonans-1-5-tesla/ has a 1.5T title and heading but describes service with a 3T device. /emar/ repeats the mismatch in a 1.5T section. The Arabic 1.5T equivalent contains an editorial correction note and raw Markdown styling. An editorial note does not establish AI authorship.

Action: operations and the medical lead approve an inventory of actual equipment and available protocols. Replace generic comparative and superiority language with accurate, supportable facts; align service pages, equipment pages and language variants.

Acceptance: field strengths, device names and availability agree across the approved inventory and public pages; editorial instructions are removed; any comparison has clinical context and approval. Open-MRI keyword mentions are not accepted as proof of an available open-MRI service.

E01 T01–T02; E04 C01–C03; E02 expanded ultrasound rendered evidence. See register F01 and F02 for exact affected URLs.

7 Make clinical accountability visible

F03 Service information lacks clear public ownership

Observed public evidence gap, P1. Sampled about and service pages use generic specialist claims without clearly identifying who examines, reports or reviews the information. English about copy gives a staff-count claim without naming the clinicians in extracted text. The coronary CT article’s structured author is an administrator account. This is not evidence that the clinic lacks qualified staff.

The useful correction is a clinical publishing system: a verified owner for each service, an approved source of operational facts and a review record for public medical content. Publish accurate professional details with the relevant individuals’ permission. Avoid implying that an author byline or schema field independently proves expertise.

A release rule the team can maintain

• A clinical owner approves procedure scope, preparation, limitations and equipment references.
• Operations approves actual availability, booking hours, supported languages and result-delivery commitments.
• A medical language reviewer checks Turkish source and complete EN/AR journeys; source approval precedes translation.
• The editor records the approval and meaningful review date, checks templates and removes drafts or editorial notes.
• The privacy/legal owner reviews claims, notices and marketing elements that require a separate decision.

Dates should mean a real review

The Regulation on Promotion and Information Activities in Health Services (Resmî Gazete, 12 November 2025, No. 33075) requires websites to state the date of the last update and contact details for the site editor (Article 5(1)(ı)). Have the clinic’s legal owner confirm the applicable presentation. Do not refresh dates mechanically: the crawl also found legal-document metadata where modification precedes publication, which warrants a CMS explanation rather than an allegation of falsified freshness.

Acceptance: the first two priority journeys have a named internal clinical approver, verified service facts, an accurate public responsibility/review presentation and a recorded next review. Every material copy change receives the same checks before it reaches all language variants.

E01 schema review; E04 C06. Regulation text: www.resmigazete.gov.tr · view source. Certificate validity and staff credentials were not audited.

8 Repair language journeys and destinations

F04 Language selection does not complete the journey

Observed, P1. /en/dental-tomography/ has English headings and an introduction followed by substantive Turkish explanation. Sampled English booking CTAs route into Turkish forms and notices. Arabic MRI pages have a different form experience. All 51 Arabic HTML records in the crawl have titles without Arabic script, largely retained Turkish titles.

Action: inventory template and language mappings, then repair the entire selected journey: body copy, navigation, headings, metadata, service selector, field labels, privacy text, validation, confirmation and human response. Give the language owner a medically approved glossary and source version. Verify that language switching reaches the equivalent service rather than a generic homepage.

Acceptance: a Turkish, English or Arabic visitor can follow each approved priority service to its intended contact route and understand the relevant notices in that language. Arabic metadata and right-to-left UI are checked by a competent reviewer. The declared hreflang relationships support these real equivalents.

F05 Working URLs can still lead to the wrong place

Observed, P1 for affected patient tasks. The crawl found shared-template label/destination mismatches on 62 HTML response records. English result-display links lead to Turkish contracted-institutions content; a laboratory navigation item leads to /en/biopsy/, whose content describes biopsy; location links can lead to the blog. The rendered Turkish ultrasound footer also sends appointment and results labels to contracted institutions.

These links return successful responses, so an HTTP broken-link report alone misses the problem. Correct the shared template from a label-to-purpose matrix and then check each language and device layout. Confirm the approved results route with clinic staff without accessing patient records.

Acceptance: appointment labels reach appointment functionality, laboratory labels reach laboratory information, location links reach practical directions and results links reach the clinic-approved entry point. No link depends on the visitor guessing which route is genuine.

E01 T03, T05 and T07; E02 browser snapshots; E04 C04. The 62-record count includes shared-template instances and is not a count of 62 unique bugs.

F18 The Turkish homepage can send visitors to the English one

Observed on 16 October 2026, P1. The root URL runs Polylang’s browser-language redirect (response header X-Redirect-By: Polylang): a browser that prefers English receives HTTP 302 to /en/home-page/. LiteSpeed Cache stores that redirect. A Turkish-language desktop Safari request received the cached 302 to the English homepage (X-LiteSpeed-Cache: hit), while uncached Turkish requests received the Turkish page. Blursor’s browser check on 15 October also ended on /en/home-page/.

One of the test requests asked for English, so it may have created that cache entry; a real English-speaking visitor has the same effect. Either way, a cached language redirect can send Turkish-speaking patients, and possibly search crawlers, to the English homepage. The Google Business Profile’s website link points to this root URL, so visitors from Maps start here too. Whether Googlebot ever receives the redirect cannot be tested from outside; inspect the homepage URL in Search Console.

Action: turn off Polylang’s browser-language detection, or keep the homepage redirect out of the cache (cache only HTTP 200 responses, or vary the cache by the language cookie). Keep the language switcher as the visible choice. Acceptance: repeated requests to / in Turkish, in English and without a language preference, on desktop and mobile, return the Turkish homepage with HTTP 200 unless the visitor has chosen another language.

E10: response headers recorded on 16 October 2026 (curl; desktop Safari, iPhone and Firefox user agents; Turkish and English Accept-Language). E05: Blursor browser ledger, 15 October 2026.

9 Technical foundations and publishing cleanup

What the crawl supports

All 186 crawl destinations ultimately returned 200. Two old internal aliases redirected once. Parsed JSON-LD had no syntax errors; there were no conflicting canonicals or unresolved reciprocal errors among the declared hreflang sets after resolving aliases. These are useful foundations to retain. They do not prove indexing, complete language coverage or eligibility for a particular search feature.

F06 Public artifacts need an owner and disposition

Observed, P2. /elementor-2653/ is sitemap-listed and indexable by its public signals, with a placeholder title and check-up package content. /preference_page/ exposes a raw shortcode. /privacy-policy/ retains WordPress sample-policy text and is linked from an English privacy/cookies label, although detailed KVKK policies also exist. /panaromik-x-ray/ has only a short label as main content. /tesekkur-ederiz/ is indexable and sitemap-listed.

Action: document the purpose and dependencies of each URL. Confirm whether package content is current and approved. Repair preference management if used. Replace or retire the sample policy through the privacy owner. Improve or consolidate the bare panoramic page. Noindex confirmation pages and remove them from XML discovery while preserving legitimate success measurement. Use 404/410 for genuinely removed content or a direct redirect to a relevant equivalent; avoid sending every retired URL to the homepage.

Acceptance: each URL has an owner and release decision; no public shortcode or placeholder remains; indexable sitemaps list intended canonical destinations; navigation and retained functionality follow the new map. Indexable signals here do not mean Google has indexed the URL.

F07 Overlap requires a service and equipment map

Observed overlap, P2. Dental service and device variants are separately self-canonical in three languages; the two English variants share a title. Arabic mammography variants coexist. Decide whether each helps a different user task. Keep useful equipment detail linked to one primary acquisition journey; consolidate only genuine redundancy after inspecting search performance and backlinks.

Acceptance: one documented primary service target per intent/language, distinguishable titles for retained pages and a reviewed redirect/internal-link/canonical/hreflang map. The crawl does not establish ranking cannibalisation or a duplicate-content penalty.

E01 T04, T10 and T11. Metadata inventory: 143 of 185 HTML response records lacked a description, 16 lacked H1, three lacked canonical. These counts include utility/legal content and redirected duplicate fetches; fix important journeys before bulk metadata work.

10 Align local business and referrer information

F08 The same organisation is described inconsistently

Observed, P2. Visible contact pages state weekdays 08:00–20:00 and Saturday 08:00–19:00. Entity JSON-LD states Monday–Saturday 08:00–19:00. The site uses the public brand Ataköy Görüntüleme while legal notices identify Solaris. The crawl also surfaces postal-code and entity-description questions. A brand and a legal entity can legitimately differ; they need a clear verified relationship. On 16 October the Business Profile showed the same hours as the visible pages (weekdays 08:00–20:00, Saturday 08:00–19:00, closed Sunday), so the JSON-LD is the record to correct.

Action: operations confirms the factual source of truth for name, legal operator, address, map pin, telephone, general opening hours, procedure-specific availability and supported languages. Align visible pages, structured data and authorised Business Profile records. Use an entity type that matches the actual authorised organisation; schema presence alone does not validate a Hospital label.

Acceptance: every representation agrees with clinic-approved facts; any special appointment hours are distinguished from reception hours; legal operator and patient-facing brand are understandable. Review profiles through an authorised owner rather than creating competing duplicate listings.

Maps decides much of local demand

A search for “emar” (about 18,000 searches a month in Türkiye, Ahrefs) shows a map pack, so a single-site imaging centre competes there before it competes in ordinary results. On 16 October the two public profiles compared as follows (E11). Ataköy Görüntüleme: 4.8 from 406 reviews; primary category “Tıbbi Tanı Amaçlı Görüntüleme Merkezi”; Monday to Friday 08:00–20:00, Saturday 08:00–19:00, closed Sunday; website link to the root URL. HSM Radyoloji at Ataköy Towers, about 90 metres away on the same E-5 side road: 4.4 from 546 reviews; the same primary category; listed as open 24 hours. The clinic’s profile is its strongest public asset: a higher rating, and recent reviews get replies, although the three most recent replies repeat the same one-line template. One review explains that the entrance is reached from the E-5 side road rather than from Ataköy itself; that belongs in the contact page’s directions. Secondary categories, listed services and profile performance need owner access, and Maps rankings were not measured. Local district pages should contain distinct, useful access information and real relevance, not near-identical doorway text.

Make the clinic easy for a referrer to use

Dental practices, gynecologists and relevant specialists may need a dependable scan/report pathway more than another generic article. The clinic should confirm whether it accepts externally referred patients, what a referrer must supply, who reports, when results are returned and which secure channel is approved. This is a service-design hypothesis; the current audit does not prove incremental referral demand.

E01 T06 and schema-review.json. Google local relevance: support.google.com · view source. The site’s existing contact location is an asset; current opening-hour truth requires owner confirmation.

11 Interpret the AI readiness check correctly

F09 Search access and training preferences are different controls

Observed controls, P2 verification. robots.txt disallows GPTBot, CCBot and Google-Extended. The observed rules do not demonstrate a blanket AI-search block. OAI-SearchBot supports ChatGPT search while GPTBot concerns potential training; Anthropic also distinguishes search, training and user-directed agents. Google-Extended is separate from Google Search inclusion and ranking.

Keep intentional training preferences unless the owner makes a different informed decision. Check search-specific robots rules, actual server/CDN responses and the Search Console generative AI inclusion setting separately. Ordinary HTTP access or a diagnostic request with a bot-like user agent does not verify a vendor crawler’s real access.

Blursor result on 15 October 2026

The requested homepage audit completed at 02:17 UTC. It returned 11 passing checks, one failing check and four informational checks, with one finding to review: the three crawler-role restrictions. The direct diagnostic received HTTP 200 and readable HTML. Browser rendering finished on the English homepage instead of the submitted root, so the tool withheld five dependent rendered-comparison checks. The saved ledger records that limitation. The redirect to the English homepage is itself a finding: see F18 in section 8. A re-run on 16 October 2026 (12:23 UTC) gave the same result: the same robots rules, the full page for every crawler-style request, and a browser check that again ended on /en/home-page/.

The raw text count was 34,451 characters. That includes repeated menus and form-option content; it is not a measure of useful clinical explanation. Some retained responses reached a capture cap, so a different-response grouping is not proof of cloaking. The tool’s headline should not be republished as a conclusion that all AI systems are blocked.

Blursor is my own diagnostic tool, publicly linked to me. I used its request ledger as additional evidence, not as independent certification. The result link can rerun the check and is not an immutable report; the dated saved result is the audit record.

What the clinic should do

Confirm the desired access policy, verify priority URLs through authorised search tools and server logs, and fix the content integrity problems that any reader or retrieval system can encounter. Missing llms.txt and llms-full.txt files are not assigned a defect score. Google’s current guide says Search ignores those files and requires no special AI schema.

E05 saved Blursor result; E06 official crawler guidance. Public tool: blursor.ai · view source. Google guide: developers.google.com · view source. OpenAI: developers.openai.com · view source. Anthropic: support.claude.com · view source.

12 Build an AI visibility baseline that can be checked

F10 Native reporting and inclusion settings need inspection

To verify, P2. Google’s dedicated Search generative AI performance report and separate inclusion control had rolled out worldwide by 31 August 2026. The old blanket statement that Search Console cannot show separate Google AI visibility is no longer current.

In authorised Search Console, inspect Settings then Search generative AI, including parent-property inheritance. Confirm priority pages are indexed and snippet-eligible as well as included through this control. Public robots.txt cannot reveal the property’s inclusion state.

The dedicated performance report documents AI Overviews and AI Mode impressions by page, country, device and date. A missing report can reflect insufficient impressions. Treat it as visibility evidence; do not promise query-level coverage, an AI-only lead count or revenue attribution from the report.

Use more than one measurement surface

Bing introduced its AI Performance preview in February 2026, expanded topic and comparison views in June and announced Topics Portfolio beta on 13 October. Availability depends on sufficient data. These reports can help identify cited pages and themes; they do not demonstrate acquired patients.

A useful prompt sample

Two fixed sets of ten Turkish questions are recorded. The Ataköy set was run in Beamtrace (ChatGPT) on 15 October 2026. The wider set asks where to get five of the clinic’s services in Bakırköy, on the European side or in Istanbul, the way patients ask (for example “Bakırköy’de sanal anjiyo yapan özel görüntüleme merkezleri hangileri?”). It was run on eight AI and search engines on 17 October; the results are in the supplement, and every answer is quoted in full, with its sources, in the AI answers document. Count a result only when the answer names Ataköy Görüntüleme or cites its website: “Ataköy” is also a district, so a district mention must not count as a clinic mention. Keep the date, the engine and model, the full question, the full answer and the cited URLs, and re-run the same sets on the same engines after release. Report each engine separately; do not add engines into one score. Do not manufacture citations or rewrite every page for a guessed prompt variant. Add English or Arabic questions only where the clinic supports the journey.

Google reporting: support.google.com · view source. Inclusion control: support.google.com · view source. Bing sources are listed in the source ledger, E06.

13 Inventory measurement by what was actually observed

F11 Several collectors are active but receipt is not equally proven

Observed, P1 review. The public site deploys GTM, GA4, Google Ads, Yandex Metrica, Microsoft Clarity, Meta and LinkedIn code. A tag in the page is different from a request, and a successful network response is different from a record in an account.

The saved evidence identifies GTM-PS5ZH8D and GA4 G-VEP2W4GEHM. A Google verification meta token is present, which supports a verification attempt; it does not establish active Search Console ownership. Legacy Universal Analytics configuration also remains in the public container. Review it during tag cleanup without claiming it created duplicate GA4 sessions.

Form and page context

Page titles and URLs were present in collection context, including medical-service paths. No patient values were entered and this audit does not prove disclosure of form contents or a medical record. Sensitive page context still deserves a deliberate minimisation and vendor-purpose review; session-replay masking should be verified rather than assumed.

Acceptance: the clinic has a vendor/tag inventory with an owner, purpose, data categories, trigger, consent rule and retention decision. Remove stale tags through an approved container release. Verify permitted data in the destination accounts and keep raw patient details out of advertising and general analytics payloads.

E02 tracking-evidence.json, supplementary-evidence.json, public GTM resource and request-summary.csv. Observation does not establish current agency ownership or justify attributing the configuration to a particular person.

14 Define a conversion before using it for decisions

F12 The appointment link is configured as an Ads conversion

Observed, P1. After rejecting cookies, an ordinary header appointment-link click navigated to /randevu-al/ and emitted a Google Ads conversion request. No appointment form was completed. This establishes a request generated by the click; account receipt, attribution and bidding use were not verified.

The public container also configures phone and WhatsApp clicks as conversions. Those actions were not exercised. A thank-you-path tag is attached to a link-click trigger while the page path contains /tesekkur-ederiz/; it is not an observed form-success event. Another information tag combines a page-load event with a clicked-element URL condition and needs review. Configuration quirks are not proof of observed firing.

Use a sequence the business can understand

A WhatsApp click is not a message, a telephone click is not a connected call, a request is not a booking and a booking is not attendance. Keep useful micro-conversions for diagnosis, clearly named and separate from the business outcome. Before relying on any Ads conversion, confirm with legal counsel whether paid campaigns are permitted at all under the 2025 regulation (section 17).

Acceptance checks

In an authorised staging or controlled test context, a booking-link click produces only its intended intent event. A successful request produces one success event after confirmed processing; failed validation and repeated confirmation visits do not create success duplicates. Test direct thank-you visits and repeat clicks. Validate the permitted events in the relevant accounts, with patient values excluded.

Use an internal request reference for deduplication only under a privacy-approved design. Do not send procedure-specific health information, free-text requests or identifiers to ad platforms merely to improve attribution.

E02 appointment-link test and public GTM trigger map. Exact diagnostic labels remain in the implementation evidence; no platform conversion count or campaign optimisation state was accessible.

15 Consent, privacy notices and form permissions

What needs review

Consent behaviour, the permissions requested on the appointment and contact forms, and the patient-results entry point need a joint technical, privacy and legal review. The specific observations (F13, F14 and the results entry point) are in a separate internal annex for the clinic’s management and its legal, privacy and IT owners. They are not reproduced here because they concern privacy, security and regulatory questions that should be handled internally.

How to test consent

Align the consent-management platform, delayed-script loader, GTM and individually embedded tags. Make intentional decisions for each category, including any necessary exempt processing, and document them. Test before choice, reject, accept, partial choice, revoke, reload and ordinary interaction across supported languages: a page-load-only tag test is insufficient when scripts wait for user interaction.

Acceptance: the network behaviour matches the owner-approved category rules in every state; rejection persists across reload and interaction; revocation changes subsequent collection appropriately; no patient data or free text enters general analytics or advertising payloads.

Form permissions and notices

Acceptance: an appointment request can proceed without optional marketing permission; notice acknowledgement and any distinct consent purposes are separate and clear; optional choices are not preselected or bundled into access to care. Verify client and server behaviour in an authorised test environment. Replace the default WordPress sample policy at /privacy-policy/ and align notice links with the approved KVKK documents (see F06).

E02 consent-state network captures and form-validation inspection. Specific observations are in the internal annex.

16 Make the enquiry experience dependable

F17 Form labels and mobile task completion need a release check

Observed sample, P2. At the mobile booking viewport, name, phone and email fields rely on placeholder text. The inspected inputs have no associated label element, aria-label or aria-labelledby. Placeholder text may provide some accessible-name fallback, but it disappears during entry and is a weak substitute for a persistent explicit label.

Action: add visible programmatic labels, clear required/optional information and meaningful validation/status feedback. Check keyboard navigation, focus, date/selector interaction, zoom and small-screen errors. Keep clinical triage and preparation questions with qualified staff instead of making an enormous selector the only way to ask for help.

The page fitted the sampled document width without horizontal overflow. The sticky contact bar and consent control occupy persistent screen space; test the submit button, validation and keyboard-open state in the actual supported devices before release. This is a targeted observation, not a complete accessibility conformance result.

Performance should be measured with the right denominator

No Core Web Vitals pass/fail verdict is assigned here. Use Search Console field groups and PageSpeed Insights/CrUX where available, then repeated mobile lab tests for representative service, article and form templates. A small clinic may lack enough page-level field data; absent data is not a pass or failure.

Inspect images, page-builder assets, delayed scripts and third-party loading only after recording reproducible metrics. Performance work must preserve consent and functional timing. Moving all tags behind a scroll listener can make an initial test look quiet without making the experience or measurement correct.

Acceptance

A user can understand and complete the approved request route on keyboard and mobile without losing labels, focus or feedback. The team records actual field and lab baselines, fixes demonstrated bottlenecks and retests the same templates. Do not promise a ranking or enquiry uplift from an unmeasured speed score.

E02 expanded mobile booking screenshot and DOM field inspection. Public performance assessment remains an implementation baseline task; no private analytics access was available.

17 What the 2025 promotion rules mean for acquisition

Paid search and sponsored social are largely closed

The 2025 regulation lets health facilities appear on search engines and social platforms only where the listing is not paid, sponsored or aimed at prominence (Art. 5(1)(i)). Sponsored promotion is allowed in the first month after opening and for Ministry-approved new medical technologies (Art. 5(1)(j)) and, under conditions, for health-tourism promotion aimed at audiences abroad with domestic targeting disabled (Art. 8). Read literally, domestic Google Ads and sponsored social campaigns are not an available growth channel. That makes organic search and the Business Profile the main digital channels, and it is why the ranking gap matters. Legal counsel should confirm how this applies to any existing campaigns and tags.

The same article bars promotion that suggests a device or service is superior to others (5(1)(h)), price, discount and campaign information (5(1)(m)) and promotional use of patient thanks or satisfaction (5(1)(e)), and requires websites to show a last-update date and editor contact (5(1)(ı)). These rules apply to organic content too: they shape what service pages may say, not only what ads may say.

What this means for the plan

Put the digital budget into the channels the regulation leaves open: accurate exam-level service pages, the Google Business Profile, referrer and insurer relationships, and measurement that shows which of these produce suitable enquiries. If the clinic wants to use an exception (the first month after a new opening, a Ministry-approved new technology, or health-tourism promotion aimed abroad), counsel should confirm eligibility first, and a campaign should start with one confirmed service, one corrected journey, permitted tracking, an authorised budget and stop criteria. No forecast or spend commitment is supplied here.

A review of the clinic’s past paid campaigns against these rules is in the internal annex. This report makes no finding that Ataköy Görüntüleme committed a particular legal breach.

Regulation text: www.resmigazete.gov.tr · view source (Articles 5 and 8). This is a textual reading, not legal advice; no exhaustive consolidated amendment audit was performed.

18 Service priorities and site architecture

F16 Repair proven journeys before fragmenting the site

Recommendation, P2. Retain the existing domain and WordPress as the starting point. Correct content, shared templates, task routes and measurement before funding a new platform or service microsites. A redesign of templates within WordPress is a separate decision from replacing the CMS. Neither guarantees more patients.

E09 Technology evidence and implementation implications

E09: user-supplied Wappalyzer export received/reviewed 15 October 2026, plus a public homepage GET returning HTTP 200. The CSV has no scan timestamp, versions or confidence values. Public Elementor markup and plugin references corroborate parts of the stack; they are not an authenticated inventory. No vulnerability or speed finding follows from product names.

Use a pilot to choose repair, template redesign or migration

• Inventory active themes/plugins, versions, licences and custom code with authorised access. Repair one representative Turkish service page and its language equivalents in staging; check clinical facts, routes, form validation, consent, translations and measured performance.
• Continue targeted repairs if the pilot meets the agreed checks at acceptable maintenance cost. Redesign shared templates within WordPress if layout or editing constraints justify it. Compare documented effort, ownership and release risk rather than choosing from a technology list.
• Consider replacing the CMS only if verified support, dependency, workflow or performance constraints cannot be resolved economically. Map every retained service URL, language equivalent, internal link, canonical and redirect before any migration; preserve useful search equity and measure the release.

Keep service priorities and authority in the decision

Choose two journeys using verified service availability, capacity and qualified demand. Give each priority examination its own accurate page, following the pattern behind HSM’s rankings (supplement): one page per examination, written around what patients search for and connected to device facts, reporting responsibility, preparation contacts, location and booking. Medically reviewed articles should support those pages. Avoid near-identical pages for synonyms, districts or guessed AI prompts. Review qualified requests and the response process before expanding investment.

Authority is part of the gap. On 16 October, Ahrefs showed DR 7 and 58 linking domains with followed links for Ataköy Görüntüleme, against DR 36 and 623 for HSM. That is not evidence of a penalty and does not justify a disavow: inspect Search Console and a representative link sample first. The useful response is links the clinic can earn legitimately, from referring physicians and dental practices, contracted insurers and institutions, local organisations and reputable directories of imaging providers. Unverified domain age or availability should not determine migration.

E01 current crawl; E03 architecture/backlink research; E09 supplied stack export and independent 15 October homepage check. Homepage: www.atakoygoruntuleme.com.tr · view source Ahrefs re-check, 16 October 2026 (E07).

Technology-detection method: www.wappalyzer.com · view source

19 A staged implementation roadmap

The timing below is a planning sequence from an agreed kickoff, not a promised delivery date or approved implementation budget. Clinical, operational and legal decisions are dependencies, not tasks to conceal inside a developer ticket.

Who needs to own what

The clinic sponsor chooses services and signs off capacity and commercial priorities. The clinical lead owns medical information. Operations owns availability and response. The editor/localisation reviewer owns publishing quality. The developer owns templates and release behaviour. The measurement owner owns event definitions and account validation. The privacy/legal owner approves purposes, notices and promotional decisions.

What can run in parallel

Clinical source correction, tag inventory and property-access preparation can proceed together. Translation follows source approval. Paid launch follows legal review, corrected journeys and measurement validation. URL consolidation follows performance/backlink review; do not treat a cosmetic URL preference as the migration decision.

All register findings have owners and acceptance checks. The proposed sequence should be adjusted to verified capacity and access at kickoff.

20 Measure outcomes by service

A compact service scorecard

Use at least an initial four-week baseline and then a comparable post-release window, extending when counts are small. Record seasonality, campaign changes, capacity, changed consent coverage and tracking repairs. A tracking correction creates a measurement break; annotate it rather than presenting it as patient growth or decline.

Economics require the clinic’s own inputs

If counsel confirms that a paid exception applies, cost per qualified enquiry equals spend divided by qualified enquiries. Expected contribution per qualified enquiry depends on the booking rate, attendance rate and contribution per attended examination. Those inputs are unknown here. Revenue is not contribution, and a model is not a forecast.

Only invest against an owner-approved allowable acquisition cost after those inputs are credible. Small samples can move sharply; report counts beside rates. No ROI figure, expected patient count or value per keyword is assigned in this audit.

Report what changed and what remains uncertain

The review should state which releases were completed, whether their acceptance checks passed, what the service-level data shows and what decision follows. Report each surface in its own units. Citations, sessions, contact clicks, accepted requests and attended examinations should remain visibly distinct.

Proposed measurement design. No first-party outcome dataset was available for this audit; the scorecard is the implementation specification, not a statement of current results.

21 Release acceptance and owner questions

Before a corrected journey goes live

• Clinical source, equipment, preparation and reporting responsibility have recorded approval; translated pages match approved meaning.
• The intended URL returns the appropriate response, canonical and snippet controls; navigation and hreflang point to real equivalents.
• Service, booking, result-entry, laboratory, location and privacy labels reach their intended destinations in each supported language.
• Visible labels, required/optional states, keyboard focus, mobile layout, validation and confirmation are usable.
• Consent choices and delayed scripts follow approved rules before choice, after reject/accept/revoke, reload and ordinary interaction.
• A contact click is distinct from a successful request; failed validation, duplicate processing and direct confirmation visits do not count as success.
• No personal values or free text appear in general analytics/advertising payloads; vendor purposes and session-replay masking have owner approval.
• A post-release crawl and authorised account checks confirm the intended state; the baseline and release date are recorded.

Questions the clinic must answer

• Which examinations are actually available now, with which devices, clinicians, languages and spare capacity?
• Which two services are the immediate business priorities, and what makes an enquiry qualified?
• Who receives each contact route, how quickly can they respond and where are booking/attendance totals recorded?
• Which legal entity operates the service, what are the approved hours/address and what is the approved secure results entry?
• Who owns the search, analytics, ads, tag-manager and business-profile properties, and which historical definitions can be trusted?
• What current legal authorisations and promotional restrictions apply to the proposed audiences, claims, channels and consent purposes?

These questions do not reduce the value of the observed findings. They prevent the implementation team from filling operational gaps with attractive copy or treating an estimated search market as booked demand.

Use the accompanying findings register as the working handoff. Mark each item open, verified, released and retested with evidence rather than replacing this dated audit with an undated status claim.

22 Evidence and source ledger

The dated local evidence archive is the record for live-site observations. Public URLs can change. The links below identify the sources used for interpretation; they do not confer endorsement of this audit.

Search and crawler documentation

Google AI optimisation guide: developers.google.com · view source

Google generative AI reporting: support.google.com · view source

Google generative AI inclusion control: support.google.com · view source

Google reporting announcement 3 June 2026: developers.google.com · view source

Google crawler controls: developers.google.com · view source

Google local ranking guidance: support.google.com · view source

OpenAI bot roles: developers.openai.com · view source

Anthropic bot roles: support.claude.com · view source

Bing AI Performance launch 10 February 2026: blogs.bing.com · view source

Bing expanded reporting 16 June 2026: blogs.bing.com · view source

Bing Topics Portfolio beta 13 October 2026: blogs.bing.com · view source

Legal sources

Regulation on Promotion and Information Activities in Health Services (Sağlık Hizmetlerinde Tanıtım ve Bilgilendirme Faaliyetleri Hakkında Yönetmelik), Resmî Gazete No. 33075, 12 November 2025: www.resmigazete.gov.tr · view source

Copy of the regulation published by the Antalya Provincial Health Directorate: antalyaism.saglik.gov.tr · view source

Digital promotion rules summary, Karabük Provincial Health Directorate (2026): karabukism.saglik.gov.tr · view source

Provider decision-information comparisons are linked in section 5, What priority service pages should answer. Public observations should be rechecked before a later implementation or publication decision.

Supplement: why HSM ranks and Ataköy Görüntüleme doesn’t

Added on 15 October and updated on 16 October 2026. These are Ahrefs model estimates for Türkiye, not actual visits, local market share or patient counts. HSM also lists a branch in Kadıköy, so its domain figures cover a wider footprint.

E07: Ahrefs Batch Analysis, 15 October 2026 (country tr; monthly volume; subdomains; both protocols); Domain Rating and referring domains from the 16 October re-check. Earlier 59/34 visit estimates are separate snapshots, not a trend.

Four reasons for the gap

• One page per examination. HSM’s search traffic comes from pages for specific examinations (3 Tesla MRI, open MRI, contrast MRI, lumbar MRI, breast ultrasound, CT angiography), each matching how patients search. Ataköy Görüntüleme mostly has general pages per imaging method, several of them sharing reused text, and its ranking terms are largely informational.
• Link authority. HSM has Domain Rating 36 and 623 linking domains with followed links; Ataköy Görüntüleme has 7 and 58. Authority isn’t everything, but a gap this size limits what content alone can achieve.
• Mixed-up copy costs rankings. The same MRI-sedation paragraph appears on the ultrasound, CT and enteroclysis pages, and Google now shows the enteroclysis page for “emar ilacı”. Pages that describe the wrong examination are hard to rank for the right one.
• Maps matters most locally. A search for “emar” (about 18,000 a month) shows a map pack, and there the clinic starts ahead: on 16 October its Business Profile showed 4.8 from 406 reviews, against 4.4 from 546 for HSM’s Ataköy Towers profile, about 90 metres away on the same E-5 side road. The gap is on the website, not on the map.

Where the traffic comes from

Ahrefs top-pages and organic-keywords reports, Türkiye, 16 October 2026. Positions change daily; the pattern matters more than any single position. The price-related terms HSM ranks for are not a recommendation to publish prices (section 17).

What to do

Correct the mixed-up copy first, because it currently teaches search engines the wrong topics. Then give each priority examination its own accurate page, keep the Business Profile consistent with the site, and earn legitimate links from referrers, insurers and local organisations. Copy HSM’s structure, not its text or claims.

Focused website and search observations only. HSM’s credentials, equipment, clinical accuracy and promotional compliance were not audited, and ranking factors beyond those shown were not measured.

AI and search visibility: eight engines, ten patient questions

E12 · On 17 October 2026 the ten wider questions (P01–P10) were put to eight AI and search engines in one run, one answer per question and engine, in Turkish, from a Türkiye context. Each engine is counted on its own; the counts are never added into one score. An answer counts for the clinic only when it names Ataköy Görüntüleme or cites atakoygoruntuleme.com.tr.

Every answer is quoted in full, with its sources, in the AI answers document that comes with this audit (HTML and PDF).

Per engine, out of ten answers: Google AI Mode 5, ChatGPT 5, Claude 3, Grok 3, Perplexity 1, Gemini 0. Google showed an AI Overview for only one question (coronary CT angiography in Istanbul), and it did not name the clinic. In ordinary Google results the clinic’s website was in the top 20 for 3 questions (positions 2, 5 and 10), and its Maps listing appeared in the map results for 2.

The clinic was named for coronary CT angiography in Bakırköy, MRI and 3 Tesla MRI, HSG and dental CT. No engine named it for the Istanbul-wide questions on coronary CT angiography (P02) or on mammography with breast ultrasound (P10), and for breast ultrasound in Bakırköy (P09) only Grok did.

Google AI Mode named the clinic through its Google Business Profile card (4.8 from 406 reviews), twice as the first centre. ChatGPT listed it in ranked lists, first once (HSG in Istanbul). The clinic’s own website was cited in 8 of the 61 AI answers.

HSM Radyoloji was named in 39 of the 61 AI answers, on six engines, and its own pages were cited in 25. The doctor directories doktortakvimi.com and doktorsitesi.com were cited in 24 and 22 answers, more often than any clinic’s site except HSM’s.

Who was named instead, and which sources were cited

P01 · “Bakırköy’de sanal anjiyo yapan özel görüntüleme merkezleri hangileri?” (Which private imaging centres in Bakırköy do virtual (CT) angiography?)

P02 · “İstanbul’da koroner BT anjiyografi için hangi merkezleri önerirsin?” (Which centres in Istanbul would you recommend for a coronary CT angiogram?)

P03 · “Bakırköy’de MR çektirmek için hangi özel merkeze gidebilirim?” (Which private centre can I go to for an MRI in Bakırköy?)

P04 · “İstanbul Avrupa Yakası’nda 3 Tesla MR olan görüntüleme merkezleri hangileri?” (Which imaging centres on Istanbul’s European side have a 3 Tesla MRI?)

P05 · “Bakırköy’de rahim filmi (HSG) çektirebileceğim yerler hangileri?” (Where in Bakırköy can I get an HSG (uterine X-ray)?)

P06 · “İstanbul’da HSG çeken özel görüntüleme merkezleri hangileri?” (Which private imaging centres in Istanbul do HSG?)

P07 · “Diş tomografisi çektirmem lazım, Bakırköy civarında nereye gidebilirim?” (I need a dental CT scan. Where can I go near Bakırköy?)

P08 · “İstanbul Avrupa Yakası’nda dental tomografi çeken merkezler hangileri?” (Which centres on Istanbul’s European side do dental CT?)

P09 · “Bakırköy’de meme ultrasonu yaptırabileceğim özel merkezler hangileri?” (Which private centres in Bakırköy can do a breast ultrasound?)

P10 · “İstanbul’da mamografi ve meme ultrasonu için iyi bir görüntüleme merkezi önerir misin?” (Can you recommend a good imaging centre in Istanbul for a mammogram and breast ultrasound?)

The separate Ataköy question set (Beamtrace, 15 October)

A separate set of ten questions about Ataköy itself, generated by Beamtrace from the clinic’s website and run on 15 October 2026 (ChatGPT only), named Ataköy Görüntüleme in nine answers, eight times first, mostly quoting its Google rating from the directory zavis.ai. Two of those questions use the clinic’s own name, so that set measures reputation rather than discovery. It is reported separately and is not added to the counts above. Beamtrace is an AI-visibility tracker I work on.

What to do

The engines that named the clinic took its facts from places where they are explicit: the Google Business Profile, and pages that state the examination. Put the place and the equipment into each priority service page (title, first paragraph, address and nearby districts), starting with breast imaging and coronary CT angiography, where no engine named the clinic for the Istanbul-wide questions. Keep the Business Profile and directory listings such as doktortakvimi.com and doktorsitesi.com accurate. Then re-run the same ten questions on the same eight engines and compare each engine separately.

Collected on 17 October 2026, one answer per question and engine, in one run. Context: Türkiye, Turkish, desktop where the route supports it, with no city location; the questions themselves name Bakırköy, the European side or Istanbul. The answers came through APIs and data providers, not through signed-in apps: Google organic results, AI Overview, AI Mode, ChatGPT and Gemini through DataForSEO; Perplexity through DataForSEO’s Perplexity API route (model sonar-pro); Claude through the Anthropic API with its standard web search tool (model claude-opus-5-5); Grok through the xAI API with search (model grok-4.7). A person using these products while signed in can see different answers: products personalise, change from day to day, and may use other models or settings. A result counts for the clinic only when the answer names Ataköy Görüntüleme or cites atakoygoruntuleme.com.tr; the district name “Ataköy” alone does not count. Raw API responses are kept for 30 days and then deleted; the answers themselves are quoted in full, with their sources, in the AI answers document, which stays part of the audit.

One answer per question and engine is a snapshot, not a share of patients or searches. Answers were not checked for medical accuracy, and some mixed up nearby centres that share an address. Claude was collected twice on 17 October: the first pass used a newer search mode that returned no citation links and, in one answer, could not search; this report uses the second pass, made with the standard web search tool. Both passes are kept with the evidence. An earlier ChatGPT run of the same ten questions, made through Beamtrace on 16 October, named the clinic in none; the model and search setting of that run are not known, so this report uses the 17 October run, where both are recorded.

Sources

HSM location and services: hsmradyoloji.com · view source

Beamtrace score definitions: community.beamtrace.com · view source

Beamtrace prompt guide: community.beamtrace.com · view source