Key takeaways
- The Smartise AI Visibility Audit is an audit contract that checks whether six AI answer surfaces (ChatGPT Search, Google AI Overviews, Google AI Mode, Gemini, Microsoft Copilot and Yandex) can reach, read and cite your website.
- The catalog holds 397 records: 93 site-readiness checks, 180 shared channel records and 124 channel-specific records, built from 247 unique check definitions.
- Every report separates completed, partial, unmeasurable and not-applicable items. A check that cannot be verified with real data is never marked as passed.
- 397 is not an official number of ranking factors, and a report never implies that 397 measurements were completed unless they were.
- The audit is delivered by Smartise, a digital growth agency in London, Ontario, so findings can be turned into fixes by the same team.
What the Smartise AI Visibility Audit is
People increasingly ask questions to systems that answer directly instead of listing ten blue links. ChatGPT Search, Google AI Overviews, Google AI Mode, Gemini, Microsoft Copilot and Yandex all assemble an answer from sources, and some of those answers name a business, quote a price, give an address or recommend a provider. If your website is not reachable by those systems, is hard to read, contradicts itself, or is confused with another business, the answer can leave you out or get you wrong.
The Smartise AI Visibility Audit is our structured way of finding out. It is an audit contract: a fixed catalog of checks, each with a defined purpose, a defined kind of evidence and a defined way of being reported. Instead of a vague promise to “optimize for AI”, the audit tells you which gate is open, which is closed, which is unknown, and who needs to act to change it.
Timing matters. An AI answer is generated at a particular moment, for a particular question, in a particular market, on a particular device and under particular conversation conditions. The same question asked an hour later, or in another country, can produce a different answer. The audit therefore records the conditions around every observation, so that a result can be understood, repeated and compared rather than treated as a permanent truth about your brand.
Why AI search visibility needs its own audit
Traditional SEO audits were designed for a results page with ranked links. That work is still valuable, and it remains the foundation: a page that cannot be crawled or indexed will not be cited anywhere. But AI answer surfaces add questions that a classic audit does not ask.
- Was the answer generated at all? Google AI Overviews do not appear for every search, so a missing citation can mean the feature never triggered.
- Which bot fetched the page? A search crawler, a training crawler and a user-initiated fetcher are different roles with different preferences, and they should not be blended into one “AI bot” setting.
- Was the source cited or merely listed? Appearing in a source panel is not the same as being tied to a sentence of the answer.
- Is the entity correct? A brand, a domain and a person can be confused with namesakes in another city or another industry.
- Is the information current and consistent? Old prices, old addresses and conflicting language versions are common reasons an answer is wrong.
- Is the measurement honest? An answer can differ between repeats, between devices and between countries, so a single screenshot proves very little.
These questions are why the catalog is split into site-level checks that apply everywhere and surface-level checks that apply to one product only. The split also stops one platform’s rules from being applied to another by accident.
The six AI answer surfaces the audit covers
Each surface has its own mechanism, its own crawler or data path, and its own limits on what can be known. The audit treats them as separate products, and it keeps their evidence separate.
| Surface | Total records | Shared | Surface-specific |
|---|---|---|---|
| ChatGPT Search | 49 | 30 | 19 |
| Google AI Overviews | 51 | 30 | 21 |
| Google AI Mode | 48 | 30 | 18 |
| Gemini | 51 | 30 | 21 |
| Microsoft Copilot | 53 | 30 | 23 |
| Yandex (Russia, yandex.ru) | 52 | 30 | 22 |
| Six surfaces | 304 | 180 | 124 |
Add the 93 site-readiness records to the 304 surface records and you reach the 397 total. The full arithmetic is explained on the 397-check catalog page.
In short, the surfaces differ like this. ChatGPT Search depends on OpenAI’s search crawler, on whether the model actually searched, and on how citations attach to sentences. Google AI Overviews and AI Mode build on Google Search eligibility, snippet controls and Search Console data, but AI Mode adds multi-turn follow-up questions. Gemini adds grounding, a verification badge and a clear split between the consumer app and the developer interface. Microsoft Copilot leans on Bing’s index, Bing Webmaster Tools and IndexNow. Yandex has its own crawlers, its own Webmaster data and an official visibility share for its Alice AI assistant. Details live on the ChatGPT and Google AI checks page and the Gemini, Copilot and Yandex checks page.
How an AI visibility audit works
An audit follows a repeatable sequence so that results from different months can be compared. The sequence is the same whether you are a single-location service business or an online shop with several language versions.
- Scope and context. We confirm your sector, your target markets and languages, the pages that matter for decisions, and who owns content, hosting and security. Scope is recorded because some checks do not apply to some sectors or page types.
- Site readiness. We run the 93 site checks: robots rules, crawler access through your firewall or CDN, HTTP responses, redirects, canonicals, indexing directives, sitemaps, rendering, structure, trust signals, entity identity, structured data and delivery.
- Question design. We build a set of decision questions that real customers ask, grouped into families. The sample, its weights and its limits are documented, because a handful of questions is a sample, not a census.
- Surface measurement. Questions are run on the surfaces that apply to your markets. For each run we record the product, the market, the language, the device and chat conditions that are known, the provider path, the raw answer, the sources and a timestamp.
- Interpretation. We separate mentions from recommendations, own-site citations from third-party citations, cited sources from merely listed ones, and neutral context from negative or false claims.
- Fix plan. Each finding is matched to a remedy that fits its cause, with an owner and a priority, rather than one generic recommendation list.
- Re-test. After fixes are shipped, the relevant checks are repeated and compared, and regressions are watched so that improvements do not quietly reverse.
What a the audit report contains
A report is organized around evidence rather than around a single score. Typical sections include the following.
Scope and coverage
Which pages, markets, languages and surfaces were in scope, how many catalog records applied, and how many were completed, partial, unmeasurable or not applicable. Coverage is shown beside every score so that nobody mistakes a narrow measurement for a broad one.
Site readiness findings
The outcome of the T01–T93 checks with the exact evidence: the robots line that allows or blocks a crawler, the status code a page returned, the canonical it declared, the structured data block that failed to parse, the price that disagrees between two language versions. Findings point to a URL and a fragment, not to a general impression.
Surface measurements
For each surface, the questions asked, the answers captured, the sources attached and the classification of each mention. Repeated runs show variability, and a zero sample is shown as unmeasured, not as zero percent.
Entity and accuracy review
Where an answer states something wrong, such as a former address, an outdated price or a confusion with a similarly named business, the report compares the claim with a dated fact approved by you and traces the passage that probably caused it.
Fix plan and re-test results
A prioritized list of changes, each tied to the findings it addresses, followed after implementation by a re-test and a regression note.
Honest status classes: completed, partial, unmeasurable, not applicable
The most important design rule in the audit is that a check which cannot be verified with real data is never counted as passed. Reports show four separate classes.
- Completed
- The evidence was obtained and the result is stated with its scope, its date and its source.
- Partial
- Some evidence was obtained, but it is limited by sample size, period, market or access. The report says exactly what is missing.
- Unmeasurable
- The provider does not disclose the information, the data path is unavailable, or the owner has not shared the needed records. Unmeasurable is a finding in itself, and it is listed rather than hidden.
- Not applicable
- The check does not fit your sector or page type, for example price checks on a page that has no prices. The reason is written down.
This approach protects you from false comfort. A tool that shows a green score for everything it could not actually test is selling reassurance, not information. A report that tells you that hosting logs are needed to verify a firewall rule, and that the logs were not provided, is more useful than one that guesses.
How the audit differs from rank trackers and classic SEO audits
The audit does not replace SEO work. It sits on top of it and asks different questions.
| Approach | Primary question | Typical output | Main limitation |
|---|---|---|---|
| Rank tracker | Where does a URL rank for a keyword? | Position over time | Says little about whether an AI answer cites you. |
| Classic SEO audit | Is the site technically sound and relevant? | Issue list and priorities | Does not observe AI answers or separate AI crawler roles. |
| Brand-mention monitor | Is the brand named in answers? | Mention counts | Often cannot tell citation from mention, or a namesake from your brand. |
| the audit | Can each AI surface reach, read, trust and cite the right entity, and how sure are we? | Evidence-based findings with coverage and status classes | Provider data is limited, so some items are reported as unmeasurable. |
The practical difference is accountability. Rank trackers report numbers because numbers are easy to collect. The audit reports what was observed, under which conditions, and with what confidence, and it refuses to turn uncertainty into a tidy percentage.
Who the audit is for
The audit is a good fit when an incorrect or missing AI answer has a real cost for your business. Typical situations include the following.
- Local and service businesses whose address, hours, service area and phone number must be right wherever an assistant quotes them.
- Clinics, practices and professional services where the relationship between a person and an organization must be described accurately and where unsupported absolute promises are risky.
- Online shops and multi-market brands with several language versions, currencies, variants and stock states that must agree with each other.
- Publishers and expertise-led brands that want their original data and expert review credited as the source of an answer.
- Agencies and in-house teams who need a defensible method for reporting AI visibility to leadership without inventing numbers.
It is less useful if you only want a single vanity score, or if you expect a guarantee that an AI system will recommend you. No audit can promise that, and Smartise does not. What an audit can do is remove avoidable barriers, correct what you control, and measure honestly what you do not.
Evidence, sources and what the audit will not claim
The catalog is built around a few principles that limit what a report is allowed to say.
- Provider facts are dated. Behaviors of search products change. When a check depends on how a product behaves today, the report cites a dated source and states the date.
- Visibility is not traffic. Being cited, being listed and receiving a visit are three separate measures. Where you share analytics data, visits and conversions from AI surfaces are reported separately from citations.
- Hidden searches stay unknown. Some systems run internal searches that the interface does not show. The audit records only what is disclosed and marks the rest as unknown.
- Ranking factor counts are not claimed. The 397 figure describes the size of an audit catalog. It is not a statement that 397 factors influence rankings.
- Training preferences are information, not score. Your choices about AI training bots are reported separately from search visibility and never change a score.
- Experimental files are informational. A file such as llms.txt is recorded for information only. It is not treated as a requirement or as a guarantee of anything.
Plain-language promise.
If the audit could not measure something, the report says so and says why. We would rather give you an honest “unmeasurable” than a confident number that cannot be defended.
Common findings and the fixes that follow
Every site is different, but a handful of causes explain a large share of poor AI visibility. The list below is a sample of patterns and the type of remedy that normally follows. Actual findings always depend on the evidence collected.
Access is blocked by an accident, not by choice
A firewall rule, a rate limiter or a bot-protection setting blocks legitimate crawlers even though robots.txt allows them. The fix is to verify genuine bot identity using official address records, allow the verified bots you want, and keep the rest restricted. This is why the audit separates robots permission from real access.
Key content only appears after scripts run
Prices, availability or the main answer are inserted by JavaScript after the page loads, or are hidden behind tabs, filters or “load more” buttons. The fix is to serve the decision-critical text in the initial HTML and make hidden sections discoverable.
The page does not answer the question it targets
A page title promises one thing and the body delivers another, the first paragraph is vague, or two pages compete to answer the same question. The fix is a clear title, a direct first answer with its conditions, and one canonical page per question.
Facts disagree across the web
Your site, your business profile and third-party directories show different addresses, hours, prices or names. The fix is a correction inventory: list each wrong record, its source and its owner, request the correction and track it until it changes.
Identity is ambiguous
A brand name is shared by another company, a person is attached to the wrong organization, or profiles do not clearly point to one entity. The fix is consistent identity information, accurate structured data and well-chosen profile links.
Old versions linger
Outdated prices survive in cached copies, PDFs, old language versions or structured data. The fix is a single source of truth, dated prices and a process for retiring old documents.
Claims have no support
Absolute promises, undated statistics and unsourced claims make a page less safe to cite. The fix is evidence: primary sources, dated figures, named reviewers and a clear description of how original data was produced.
What we need from you, and why
An audit is faster and more accurate when you can share a few things. None of them is required to begin, and anything you cannot provide simply shifts the related checks to partial or unmeasurable, with the reason noted.
- Server access logs or a log export. These are the only reliable way to confirm that a verified crawler actually reached your pages, and to tell a firewall block from a robots rule.
- Search Console and Bing Webmaster Tools access. Owner data about index status, canonical selection and AI-related reports is far more dependable than anything inferred from outside.
- Analytics access. Visits and conversions that arrive from AI products let us separate real traffic from citations.
- A fact sheet. The correct legal name, brand names, addresses, phone numbers, hours, service areas, current prices with their dates, and the person responsible for keeping each page up to date.
- A list of the questions customers ask. Your team knows the real decision questions better than any keyword tool.
- A list of what must never be said. Claims you cannot support, regulated wording and sensitive topics help us interpret negative or false statements correctly.
We handle this information carefully and use it only for the audit. If you prefer not to share a category, say so; the report will state which checks were therefore limited.
Working with Smartise
Smartise is a digital growth agency based in London, Ontario. The team designs and builds websites, including WordPress and e-commerce sites, and runs search engine optimization, content marketing, advertising and brand work. The audit connects that practical experience to the new question of AI answers: the same people who audit your visibility can implement the fixes, so findings do not sit in a report that nobody acts on.
A typical engagement starts with a short conversation about your business, your markets and the questions your customers ask before they buy. From there we agree the scope, the surfaces and the pages that matter. If you want to move quickly, begin with the 93 site-readiness checks, which improve your standing in classic search as well as in AI answers. If you want to understand the method first, read the catalog explanation and the AI search visibility guide.
To discuss an audit, contact Smartise or call +1 519 494 3673 during business hours (Monday to Friday, 9:00 am to 7:00 pm Eastern).
Frequently asked questions about the audit
What is the Smartise AI Visibility Audit?
The Smartise AI Visibility Audit is an audit contract that checks whether ChatGPT Search, Google AI Overviews, Google AI Mode, Gemini, Microsoft Copilot and Yandex can reach, read and cite your website. It consists of 397 catalog records covering site readiness and surface-specific measurements.
Does 397 mean there are 397 ranking factors?
No. The number 397 is the size of the audit catalog. It is not an official count of ranking factors, and a report does not claim that 397 measurements were completed. Each report separates completed, partial, unmeasurable and not-applicable items.
Can the audit guarantee that an AI system will recommend my business?
No. No audit can guarantee a recommendation, because AI answers vary by question, market, device and moment. The audit removes avoidable barriers, corrects what you control and measures the rest honestly.
How is the audit different from a rank tracker?
A rank tracker reports the position of a URL for a keyword. The audit checks whether each AI surface can reach your pages, whether it cites them or only lists them, whether it identifies your business correctly, and how confident the measurement is.
What happens when something cannot be measured?
It is reported as unmeasurable with the reason, for example that a provider does not disclose the data or that server logs were not shared. A check that cannot be verified with real data is never marked as passed.
Which AI surfaces are included?
Six surfaces are included: ChatGPT Search, Google AI Overviews, Google AI Mode, Gemini, Microsoft Copilot and Yandex (Russia, yandex.ru). Each has its own surface-specific checks plus 30 shared measurement rules.
Do I need to be an SEO expert to use the findings?
No. Findings are written in plain language, point to the exact page and evidence, and come with an owner and a priority. Smartise can also implement the fixes for you.
Is llms.txt required for AI visibility?
No. The audit records the presence of llms.txt for information only. It is an experimental file and is not treated as a requirement or a guarantee. Crawl access, indexing, clear content and consistent identity matter far more.
Continue exploring the AI visibility audit
397-Check Audit Catalog
How 397 records break down, how to read the numbers and the 30 shared measurement rules.
Read more about 397-Check Audit Catalog →Site Readiness Checks
All 93 site-level checks, grouped by crawl access, indexing, structure, trust and identity.
Read more about Site Readiness Checks →ChatGPT, AI Overviews & AI Mode Checks
The 58 checks specific to ChatGPT Search, Google AI Overviews and Google AI Mode.
Read more about ChatGPT, AI Overviews & AI Mode Checks →Gemini, Copilot & Yandex Checks
The 66 checks specific to Gemini, Microsoft Copilot and Yandex (Russia).
Read more about Gemini, Copilot & Yandex Checks →AI Search Visibility (GEO) Guide
A practical, hype-free guide to being found, read and cited by AI answer systems.
Read more about AI Search Visibility (GEO) Guide →Ready to grow with Smartise?
Tell us about your business and goals. We will recommend the smallest set of changes that makes the biggest difference, and tell you plainly what we can and cannot promise.