To understand how AI tools describe your business, examine real customer questions, the conditions of each check and the sources behind the answer. Asking only for your company name gives an incomplete picture. A useful review separates accurate information, citations and recommendations, then identifies practical website improvements and a clear route to a relevant enquiry.
Define the question behind the visibility request
“Does ChatGPT know our business?” is understandable, but it is difficult to turn into a sensible scope of work. Better questions are whether a tool describes your services accurately, confuses you with another company or links to a page that helps a prospective customer make a decision. These can be investigated and recorded.
A mention, citation, recommendation, visit and qualified enquiry are separate observations. An answer might describe a company correctly without linking to it. Another might cite a useful guide without recommending the publisher as a supplier. Reporting should preserve those distinctions rather than bundle them into a single success score.
The 2027 planning perspective here describes a process that can be prepared in November 2026. It is not a report of future performance or a prediction of unreleased product features. Check the current interface and official documentation when carrying out the work, and date the evidence you collect.
| Observation | Evidence | What it does not establish |
|---|---|---|
| Mention | Exact answer and observation conditions | Recommendation |
| Citation | A working source link | Every user sees the same source |
| Enquiry | Accepted submission or recorded contact | A sale |
How should a visibility study of Turkish brands be planned?
Define the selection method before choosing businesses. Record the sector, service type, language and market; familiar or already-visible companies do not represent every Turkish brand. This is a method for planning a 2027 review, not a completed 2027 study or a WeAreMedia client result.
Keep three question groups: company identity, customer needs without the brand name, and supplier-selection criteria. Asking where a named company operates differs from asking what a retailer should prepare before migrating a store. Neither should be silently counted as an independent recommendation. Use genuine public questions without transferring private customer conversations.
Record the sample size and selection rules in advance. Including several sectors does not make the sample statistically representative. Keep unavailable searches and unassessed sources separate from valid observations without a mention. Explain the denominator before reporting a proportion.
For each observation, record the question, date, language, visible mode, answer and opened sources. Separate mentions, citations and recommendations. ChatGPT’s source guidance and Gemini’s related-link guidance support checking the underlying evidence, rather than trusting an icon.
Publish the actual observation period, question set and limits. Do not combine unlike sectors into one success percentage or claim growth after changing the selection method. Until observations exist, label the deliverable a research plan, with no invented findings.
Make the observation understandable to another person
Record the date, language, country specified in the question, tool, visible mode or model and whether web search was used. Start a fresh conversation when earlier discussion could influence the answer. This reduces one source of variation, but does not establish that every form of personalisation has disappeared.
Keep the exact question. Separate fields for company name, service description, location, contact information and cited page make the result easier to assess. Preserve a short excerpt or necessary screenshot where it helps substantiate a finding. Avoid collecting complete conversation histories when a small evidence record will do.
Include a “not assessed” status. If a tool could not search, the observation should not become a zero in a visibility calculation. A blocked feature or regional access limit is different from an available search that did not mention the business. Explain the denominator before presenting any proportion.
Separate discovery from reading supplied information
Giving a tool your website and a description of every service before asking what the company does tests its use of supplied context. It does not establish independent discovery. That can still be a useful readability exercise, but label it separately in the assessment.
First ask a customer-need question without providing your own page. In a second pass, supply the address if you want to test how accurately the information is understood. A difference might support a limited finding: the page was understood when provided, but was not independently selected in this observation. It does not, by itself, establish why that happened or prove a technical block.
Keep the reviewer’s interpretation separate too. Familiarity with the company should not cause a missing service to be counted as correctly described. A clear record helps another reviewer assess the same evidence without assuming the intended outcome. A manageable observation sheet can be enough to begin; an expensive score does not substitute for a transparent method.
- Choose a customer question
- Record answer and conditions
- Open the cited source
- Correct the verified fact
Open the source rather than trusting the citation label
OpenAI’s search guidance notes that results and citations can be incomplete, outdated or incorrect. Open the linked page and check whether it supports the particular claim you are assessing. A citation icon is a starting point for verification, rather than proof that every sentence is accurate. Official ChatGPT search guidance.
Claude’s web-search documentation describes responses with citations that can be checked. Review the page behind an attribution and distinguish what it states from what the answer infers. A source can support one part of a response without supporting a broader business claim. Claude web-search guidance.
Pay particular attention to discontinued services, old addresses and work belonging to another company. A tool saying that a fact came from your website is not sufficient evidence. If the cited page does not contain or support it, classify the statement as unsupported by that source.
What should you do when Gemini gives the wrong service or location?
Reduce the concern to one checkable claim, such as “this company has an office in London”. Record the question, date, response language and visible search conditions. Open the company’s contact page. Serving customers in a country does not establish an office there. This is a hypothetical example, rather than an observed Gemini answer about WeAreMedia.
If the answer provides a source, open it and look for support for that particular claim. A page mentioning the company is insufficient evidence for an office location. Gemini does not show sources for every response; an absent source is not proof that your website is blocked. Mark the evidence as supported, unsupported or not checked. Google’s source explanation sets out the relevant limits.
If your own page is wrong, verify the facts with the business before correcting it. If the source is accurate but the answer makes an unsupported inference, do not invent an office page to match it. Consider the response feedback control. Before sending anything, review what associated conversation and attachments may be shared, and exclude private customer information. Google’s feedback guidance explains that related conversation content can be included. Feedback is not a promise to change answers for every user.
Record the corrected page and live verification separately from the next answer observation. Reuse the recorded question under comparable conditions. If the answer remains unchanged, keep the finding open rather than repeatedly refreshing publication dates. Our SEO service can help investigate the underlying information and source problem. A website address and a short example without personal data are enough for an initial discussion.
Eight checks when ChatGPT does not recommend your business
One missing recommendation does not establish a known fault. Use these checks to investigate possibilities rather than claim a diagnosis. OpenAI’s search guidance does not guarantee placement.
- Question: Is the user choosing a supplier or learning a method? An explanatory answer without an agency recommendation is not automatically lost commercial visibility.
- Conditions: Was web search used, and what language, location and prior context applied? A response without search is not a comparable test of current web-source selection.
- Supplied information: Did you put the company and its address into the question? That tests supplied context. Keep it separate from a customer question that does not name the business.
- Crawler access: Review the relevant route and access layers for OAI-SearchBot. GPTBot has a different training role; blocking it alone does not establish a search block. Official crawler roles.
- Owning page: Can someone understand the service, intended customer and enquiry route on one appropriate page? Improve that page before creating another address competing for the same need.
- Identity: Do service, company and contact pages agree? Verify the facts with the business. Serving overseas customers does not establish an overseas office.
- Source support: Does the cited page substantiate the claim? Correct outdated information you control; prepare evidence for external corrections. Another publisher’s update schedule is outside your control.
- Outcome: Record mentions, links, recommendations, visits and enquiries separately. Keep the question stable when rechecking. One changed answer does not establish market-wide growth or prove causation.
The review should finish with a specific page problem, evidence and a bounded task. We can use this checklist when discussing SEO work with WeAreMedia, agreeing the issue before proposing implementation.
Improve the page that should answer the customer
A commercial page needs to explain who the service suits, what can be included, which inputs are required for a proposal and where the limits lie. Clear responsibilities and deliverables help someone choose a supplier. Hidden instructions asking an AI system to recommend you do not provide that decision support.
A guide has a different role. It should explain a method, its conditions and a practical example. Instead of expanding the statement that SEO matters, show when a business should investigate technical access before commissioning more content. Our guide to content for Google and AI search discusses how these needs fit together.
Evidence should sit close to the claim it supports. If a real project screen is shown, explain the agency’s actual role. Do not invent customer totals, conversion improvements or research samples. When using someone else’s study, name the source and coverage rather than presenting it as your own work.
An illustrative review of a confusing wholesale offer
Imagine a manufacturer with both wholesale activity and a retail store. An AI answer says wholesale orders can only be placed through the retail checkout. Before repeatedly asking for a different answer, identify where the site explains the wholesale process. This example is hypothetical, with no implied customer result.
If delivery coverage, order conditions and the quotation route are unclear, obtain the correct information from the business and improve the appropriate page. Add a relevant link from the retail experience to the wholesale section. Do not invent a distributor network or a guaranteed delivery date to make the explanation sound complete.
Repeat the observation with the recorded question and conditions. Even if the answer improves, that alone does not demonstrate that one page change caused it. External sources and product changes may also matter. The useful outcome of the exercise is an evidenced problem, a justified implementation and an honest account of what remains uncertain.
Measure a defined sample without turning it into market share
You can report how many valid observations were made and how many included a mention or citation within a fixed question set. That proportion belongs to the selected sample. It does not estimate the probability that every customer will see your company, nor your share of the wider AI market.
If the question set changes, do not calculate a straightforward growth percentage against the old one. Keep a stable core and label any exploratory additions separately. A repeated review should help identify persistent errors and useful changes, not reward a team for choosing easier questions after the first report.
Website visits belong in a separate evidence stream. Record AI referrals when they genuinely appear in Analytics. Their absence does not prove that nobody arrived after using an assistant. Consent, application behaviour and link handling can limit attribution. Do not add answer observations to measured website sessions as if they were the same unit.
A telephone or WhatsApp click signals intent, rather than a completed conversation. A successful form submission, a suitable enquiry and a sale are also different events. Our measurement setup service can help define that reporting scope. Personal information and message contents should stay out of Analytics.
Turn a finding into a bounded task
Use four fields: the specific problem, supporting evidence, proposed intervention and completion check. For example, if an old telephone number remains on your own contact page, verify the correct number and update that page. The immediate completion check is the live information and telephone link. An assistant updating its answer is a separate later observation, rather than a deliverable within your control.
This distinction makes the scope reviewable. A statement such as “GEO completed” does not explain which fact or experience changed. A record showing the affected address, the evidence and the implemented correction is much more useful for the business and the next reviewer.
Decide how much work the finding justifies
An initial engagement can be a bounded review: priority services, question set, source verification and an implementation list. Copy, interface changes, development and measurement can then be scoped separately. Running more prompts every month has limited value if the main website problems remain unresolved.
Languages, service range, inconsistent legacy content and technical dependencies affect the effort. For several markets, translating the same question is insufficient. Consider what that audience actually needs and which regions the business genuinely serves. An unsupported overseas presence should never be introduced as an optimisation tactic.
Ask for an accountable scope. Each recommendation should name the page, owner, reason and verification method. If the finding is uncertain, it should remain uncertain in the proposal. A review is not completed implementation, and completed implementation is not proof of a commercial improvement.
Discuss the actual website problem with WeAreMedia
Share your web address, priority service and an example of the description that concerns you. We can discuss SEO and organic visibility through technical access, service information, content and the enquiry journey. An introductory message does not require administrator passwords or customer lists.
Call to discuss how your business is described, send your site and priority on WhatsApp or use the short contact form. First establish the verifiable issue, then agree the content or implementation needed to address it.
Questions to settle before the first check
Does one missing mention establish a technical problem?
No. Wording, context, tool and source selection can differ. Record limited observations under comparable conditions, and investigate access or indexing through separate evidence.
Can repeating a company name guarantee recommendations?
No. Prioritise accurate service information, useful answers and a usable page. Source and recommendation choices remain outside the business’s control.
What should the assessment deliver?
A useful record contains the questions, observation conditions, verified inaccuracies, source URLs, priorities and checks. It should distinguish preparation, implementation and commercial outcomes, making clear what has actually been established.
