To check AI visibility, choose ten real buyer questions, run each across five selected AI search experiences, and save whether your business is mentioned or linked. That creates 50 checks in one baseline run. A second complete run adds another 50 responses. Record unavailable answers separately and use the same questions and conditions when you compare results.
AI visibility describes whether a business appears accurately in generated answers and whether those answers cite its website. It is a sampled observation, not a universal ranking or a promise that any person will see the same response.
How SEO, AEO and GEO fit together
SEO helps search engines discover and understand useful pages. AEO focuses on clear answers to people’s questions. GEO concerns visibility in generated answers. They overlap: accessible pages, relevant information, a clear business identity and evidence all help people and search systems understand a site.
Google says its established SEO guidance also applies to its AI search features; there is no separate AI markup requirement or special file you must add. Read its guide to optimising for generative AI features. Structured data should describe the visible page accurately, rather than promise an AI citation.
Choose questions that match a buying decision
Use sales enquiries, customer conversations and keyword research to choose your prompts. Include different decisions: finding a provider, comparing alternatives, understanding price and checking suitability. For example:
- Which [service] providers work with [customer type] in [market]?
- What should I check before hiring a [service] provider?
- How much does [service] cost, and what is usually included?
- Which providers offer [specific requirement]?
Keep prompts that name your own business in a separate brand-accuracy check. They can reveal incorrect details, but mixing them into discovery tests inflates the chance of a mention. Keep a stable core question set; if you add questions next month, report the new set separately.
Run a reproducible baseline
Choose the experiences your buyers use. The Fix Kit’s default rows cover ChatGPT Search, Perplexity, Gemini, Google AI Overviews / AI Mode, and Claude. You can substitute another experience, such as Copilot, if relevant, but record the change. Google AI Overviews and AI Mode are distinct experiences: record which you used rather than switching silently within a comparison.
- Record the date, market, exact prompt, platform and available model or mode.
- Use a fresh conversation and document any account, memory or personalisation settings. A fresh chat does not remove every source of variation.
- Save the actual response, source URLs and an evidence reference where permitted.
- Record whether a generated answer appeared. If it did not, mark it as unavailable or not triggered rather than inventing an answer.
- Run all ten questions once across the five selected experiences. Repeat the complete set separately if you want a second sample.
A baseline requires manual testing and review. Platform availability, account access and usage limits can affect the time needed. Fifty checks is a small sample of the questions real buyers could ask.
Measure mentions and citations separately
| Metric | Definition |
|---|---|
| Mention rate | Usable answers naming your business divided by usable answers in the defined sample. |
| Website citation rate | Usable answers linking to your website divided by usable answers in the same sample. |
| Answer coverage | Usable answers divided by all planned checks, with unavailable results explained. |
| Brand accuracy | Whether the response correctly describes your services, location and other verifiable facts. |
Also record competitors and sources. A mention without a website link is not a website citation. A directory citation about your business is another useful observation, but should not be quietly counted as a link to your own site. If you use a composite score out of 100, publish its weights and treat it as an internal comparison measure, not an official score assigned by an AI platform.
The worked example shows the arithmetic and explains why the denominator matters.
Investigate access using the right crawler
Search access, model training and user-requested page fetching are different activities. OpenAI documents OAI-SearchBot for search discovery, GPTBot for training-related crawling, and ChatGPT-User for user-initiated requests. Search and training preferences can be controlled separately. Use the current OpenAI crawler documentation, including its IP verification guidance, when reviewing server rules.
For Google Search, review Googlebot access, indexing directives and search controls. Google-Extended does not control inclusion or ranking in Google Search; it governs specified training and grounding uses. Google explains the distinction in its crawler documentation.
Check the actual robots.txt rules, HTTP status, canonical URL, robots meta tag and any firewall or CDN challenges. A robots.txt allowance alone does not prove that a crawler received the content. Use verified crawler logs where available. Conversely, a missing mention does not establish that access is blocked.
Improve pages using evidence
For each important service or product, make it easy to find what it is, who it suits, what it includes, how the process works, what it costs and what evidence supports the claims. Start with a direct answer, then add the detail a buyer needs. Use descriptive headings and internal links that lead to the supporting pages.
Make the organisation name, founder, contact details and business location consistent. Use accurate Organization, Person, Article, Product, Service or breadcrumb markup where it describes the page. Do not invent reviews, ratings, clients or locations. Keep useful visible FAQs, but do not build a strategy around a guaranteed rich result or AI quotation.
When repeated answers cite a directory or publication, investigate why that source is relevant. Correct inaccurate profiles and earn useful coverage through real work. Buying irrelevant mentions or publishing large numbers of repetitive pages is not a substitute for useful information.
Track outcomes beyond a manual prompt test
Track identifiable AI referral sessions in GA4 alongside qualified enquiries or sales, with a consistent source definition. Referral data is incomplete: some answers produce no click or do not pass a usable referrer. Separate observed referrals from assumptions about untracked influence.
Use Google’s Generative AI performance report where available for the property. Check the current Search generative AI setting as part of the account review. These reports and controls complement manual testing; they do not describe every assistant or prove that a particular edit caused a change.
Branded search trends can provide context, but a rise in brand queries is not proof of AI influence. Advertising, recommendations, offline activity and other exposure can also change demand. For reliable comparisons, first verify the conversion events.
Common questions
Can anyone guarantee that ChatGPT will recommend my business?
No. A website owner can improve access, clarity and evidence, but cannot guarantee an organic mention or citation in a generated answer. Paid advertising and organic citations should be evaluated separately.
How often should I repeat the checks?
A monthly baseline is a practical starting cadence for a small business. Repeat sooner after a major site change or a suspected access problem, and record why the schedule changed. Choose a cadence that supports an actual decision rather than collecting numbers without acting on them.
How long does improvement take?
There is no dependable universal timeframe. Discovery, recrawling, indexing, retrieval and generated responses can change at different rates. Record the implementation date, verify the technical change first, and compare repeated samples without promising a four-to-eight-week result.
Do I need llms.txt?
It is not a requirement for Google’s AI search features. Prioritise accessible HTML, useful pages, clear identity, valid metadata and sound measurement. Evaluate any additional file against documented support and a concrete use case.
Choose a workbook or a done-for-you audit
The AI Visibility Fix Kit costs $9.89 USD at the introductory price. The download includes an Excel workbook with 100 prompt templates to adapt manually, 50 test rows, readiness checks, competitor comparison, a dashboard and a 30-day plan, plus the supporting guides and templates listed on the product page. It does not automatically run the tests or edit your website.
The AI Visibility Audit costs $220 USD for the first audit, with a standard rate of $290. It includes up to 50 checks across five agreed experiences, an access and content review, and a written prioritised plan. Standard delivery is five working days after the intake and access are complete; implementation is quoted separately.
For ongoing work, see SEO, AEO and GEO services, or send a question about your site.