An AI search visibility audit checks whether an AI system can find your page, understand it, and describe it accurately. Those are three separate conditions, and most sites fail at least one without knowing it.
The audit below takes about one afternoon and needs no paid tool. It covers seven checks, what a failure looks like, and the specific fix for each. It also covers the part most checkers skip, which is what you cannot control no matter how well you optimise.
What does visible actually mean?
In an ai search visibility audit, visible means three things in sequence, and a page has to pass all three.
The system retrieves the page. That requires crawler access and an indexable page. If a bot is blocked or the page is noindexed, nothing else matters.
The system cites the page. That requires a passage that is useful, specific and distinctive enough to be worth attributing. Plenty of pages get retrieved and absorbed without a credit.
The system describes the page correctly. That requires the page to state plainly what it is, who made it and what it covers. A page that is vague about its own subject gets summarised vaguely, and a vague summary is worse than no mention because it is wrong in public.
An ai search visibility audit that only checks the first condition is a technical crawl check. A complete audit checks all three, and the third is where most sites lose the most.
The three failure modes
Each condition has a matching failure, and each failure needs a different fix from an ai search visibility audit.
| Failure | What it looks like | Usual cause |
|---|---|---|
| Not retrieved | The page never appears in AI answers at all | Crawler blocked, noindex, or thin content |
| Retrieved, not cited | Your information appears, credited elsewhere | Passages that are useful but not distinctive |
| Cited, described wrong | You appear, described inaccurately | Vague self-description near the top of the page |
The second failure is the hardest to notice during an ai search visibility audit, because there is no click to miss. You cannot see it from your own analytics, because there is no click to miss. It shows up as competitors being mentioned for a fact you published first, and it is the most common outcome for pages that are well written but generically phrased.
The third failure is the most damaging outcome an ai search visibility audit can find. A page that never appears is invisible. A page that appears with the wrong description is actively misleading, and the correction takes months because the system has to re-crawl and re-summarise.

The seven checks
Run the seven checks in this order. An ai search visibility audit is sequential, and the early checks invalidate everything after them.
Check 1: crawler access. Fetch your robots.txt and confirm the citation-friendly bots are not blocked. Google documents its own crawlers in the crawler overview, and OpenAI documents its bots at platform.openai.com/docs/bots. If a major bot is disallowed, you have removed yourself from the pool before anyone assessed your content.
Check 2: indexability and canonical. Confirm the page returns 200, is not noindexed, and its canonical points to itself. A page with a canonical pointing elsewhere is telling the system that a different URL is the real one, and the system will believe it.
Check 3: entity clarity. This is the check an ai search visibility audit most often finds failing. Read the first hundred words. Do they state what this page is, who publishes it and what it covers? If the answer requires scrolling, the summary will be built from whatever appears first, which may be a navigation menu.
Check 4: chunk independence. Take each H2 and its section, paste it into a document alone, and read it. If it needs the section above to make sense, the chunk is not independent and it will be retrieved poorly.
Check 5: claim specificity. Count the numbers, dates and named entities on the page. A page with none is hard to cite because there is nothing quotable. Three specific, checkable facts do more than a thousand words of description.
Check 6: primary sources. Confirm the outbound links go to primary sources rather than to other blogs summarising them. Citing the original is a trust signal. Citing a summary of the original is a signal that you did not read it.
Check 7: freshness. The last check, and the easiest one to fix once the others pass. Look for a visible date and a genuine update. Content refreshed within the last few months is treated differently from content that has not changed in three years, and the difference is visible on the page rather than only in the database.

How to run each check without a tool
Every check in an ai search visibility audit has a manual version. The table below is the whole audit, compressed.
| Check | How to run it | Pass condition |
|---|---|---|
| Crawler access | Open yourdomain.com/robots.txt | No Disallow for major AI bots |
| Indexability | Open the page, view source, check the robots meta | 200 response, no noindex |
| Canonical | View source, find the canonical tag | Points to the page itself |
| Entity clarity | Read the first hundred words | Subject, publisher and scope stated |
| Chunk test | Paste one H2 section into a blank doc | Reads as a complete answer |
| Specificity | Count numbers, dates and names | At least three checkable facts |
| Sources | Review every outbound link | Primary sources, no broken links |
That is a four-hour job for a site with fifty pages, and about forty minutes for a single article, which makes an ai search visibility audit the cheapest technical review available. The chunk test takes the longest, because it is the one that requires reading rather than checking.
There is a shortcut worth knowing when you run an ai search visibility audit on a large site. Run the chunk test first on your three highest-traffic pages. If those pass, the rest of the site usually follows the same pattern. If they fail, the fix is structural and it applies everywhere.
The fix table
Each failure found by an ai search visibility audit has a specific fix. Generic advice like “improve quality” is not one.
| Check | Symptom | Fix |
|---|---|---|
| Crawler access | Bot receives a 403 or a Disallow | Allow the citation bots in robots.txt |
| Entity clarity | AI misnames the product or the author | Name both in the first sentence |
| Chunk test | Section needs the one above it | Restate the subject at the start |
| Specificity | Vague claims only | Add a number, a date or a name |
| Sources | Links point to summaries | Replace with the primary source |
| Freshness | No visible date | Add a date and a real changelog |
Two of these fixes are one-line changes and two are structural, and an ai search visibility audit is only useful if it ends in one of them. The structural ones, entity clarity and chunk independence, are worth doing first because they improve the page for readers at the same time.

Crawler access, in detail
Crawler access is the check that invalidates everything else in an ai search visibility audit, so it deserves its own section.
Three groups of crawlers matter to any ai search visibility audit. Search crawlers that feed the classic index, AI crawlers that feed training and retrieval, and answer-engine crawlers that fetch pages to cite them in a live response. Blocking the first group removes you from search. Blocking the third removes you from citations.
The practical rule is to allow the citation bots and to think carefully about the training-only bots. A bot that trains a model without ever linking back is taking content and returning nothing measurable. Blocking it is a business decision rather than a technical one, and it is defensible.
Two mistakes show up repeatedly when an ai search visibility audit reaches this step. Blocking every AI bot in a single line during a security scare, and then wondering why citations dropped six months later. And allowing everything without checking, which means a staging site or a duplicate domain can end up competing with the real one.
Check the file rather than the dashboard. Robots.txt is a plain text file and it takes ten seconds to read in full. Most audit tools summarise it, and summaries miss the specific user-agent lines that decide whether a bot is allowed in, which is the whole point of the check.
What you cannot control
No ai search visibility audit controls whether an AI system cites you. This is the honest limit, and pretending otherwise is how visibility tools end up making promises they cannot keep.
| What you control | What you do not control |
|---|---|
| Whether bots can fetch the page | Whether an AI cites you |
| Whether each section stands alone | Which competitor it prefers |
| Whether claims are checkable | How it summarises you |
| Whether sources are primary | When it refreshes its index |
The right frame for an ai search visibility audit is eligibility, and eligibility is the part generative engine optimization actually controls. An ai search visibility audit makes a page eligible for citation. It cannot make citation happen, and it cannot make it happen for a specific query on a specific day.
That limit is a reason to define what an ai search visibility audit is for. Eligibility is a precondition, and a page that fails the seven checks is not in the pool at all. The audit moves you from “cannot be cited” to “could be cited”, which is the only part of the process that responds to effort.

Reading the result honestly
An ai search visibility audit produces a list of pass and fail results, and the temptation is to treat a pass as proof of performance.
It is not. A pass means the page is eligible, which is necessary and not sufficient. The useful reading is comparative: which pages pass, which fail, and whether the failures cluster in one section of the site. A cluster usually points at a template problem rather than a writing problem, and a template fix is cheaper than fifty rewrites.
The second useful reading is temporal. One audit is a snapshot. Three audits show whether a fix held, whether a crawler rule got reverted, and whether a competitor overtook you on a query you used to own.
Record the date, the seven results and the fixes. That log is the only durable output of an ai search visibility audit, and it is what turns the exercise into a system rather than an afternoon.
What to fix first when everything fails
If several checks fail at once, the order is not arbitrary. Access first, then indexability, then the content checks. Fixing a heading before a crawler rule is wasted effort, because the page is not in the pool yet.
The second rule is to fix templates rather than pages. If every article on the site fails the chunk test, the problem is the template that renders the opening block, and one change fixes fifty pages. If one article fails, the problem is that article.
The third rule is to record what you changed and when. Crawler rules get edited during incidents and restored months later, or never. A short log with the date, the change and the reason turns a recurring mystery into a known history.
There is a fourth rule that saves the most time. Do not chase every failed check on every page. Pick the ten pages that carry the most traffic or revenue, fix those completely, and leave the rest until the next cycle. A perfect long tail is worth less than ten pages that work.
Most sites do not need a bigger review. They need to finish the one they started, and to write down what they changed so the next person does not undo it by accident. That single habit turns a one-off project into something the site keeps, and it costs about four lines in a shared document that nobody has to maintain.
The llms.txt question
A file at the root called llms.txt, described at llmstxt.org, is a markdown summary of a site written for AI systems rather than for search crawlers. It is a low-cost addition and it is not a priority.
The position in 2026 is that adoption is uneven, and the audit should record the file without weighting it. Some systems fetch it, most do not, and there is no reliable way to measure the effect. Adding one takes twenty minutes and costs nothing. Expecting it to change your visibility is a different claim, and the evidence does not support it yet.
Treat it as one line in the checklist rather than a strategy. If it is there, good. If it is missing, it is not the reason a page is not being cited.
Where the engine fits
The engine supports the content half of an ai search visibility audit rather than the technical half. It collects evidence, groups gaps by type and attaches an evidence class to every claim, which is the raw material checks three to six depend on.
In a live run on 7 November 2026, the engine produced 54 gap records and 15 opportunities for one topic, with each gap typed as a content, question, evidence, trust or AEO gap. The AEO gaps are the ones an ai search visibility audit cares about, because they mark the questions where no page in the result set gives a direct, citable answer.
Two outputs map onto the audit.
- Evidence gaps. A gap typed as evidence means a claim exists without a source. That is check six, surfaced before you publish rather than after.
- AEO answer blocks. The report includes a direct answer section written to be quotable, which is the output of checks three, four and five combined.
The AEO gap type and the quotable answer block are documented on the Content Gap Finding Engine page, with the reporting layer at SOVEL Gap Brain.
The structural argument sits in query fan-out, which explains why chunk independence matters mechanically. For the writing layer, see LLM content optimization and answer engine optimization for business websites.
Running the audit on a schedule
Once a quarter is enough for most sites, and once a month for a site publishing weekly. The audit is a habit rather than a project.
Three things change between audits. Crawler rules get edited during incidents and never restored. Pages get refreshed without updating the visible date. And competitors publish pages that pass the checks, which changes what the system has to choose from.
Keep a simple log with the date, the seven pass or fail results, and the fixes applied. After three audits you have a trend rather than a snapshot, and a trend is what tells you whether the work is having any effect.
FAQ
How long does an AI search visibility audit take?
About four hours for a fifty-page site and forty minutes for a single article. The chunk test takes most of that time, because it requires reading each section in isolation rather than checking a tag.
Do I need a paid tool to run one?
No. Every check has a manual version using the page source, the robots file and a blank document. Paid tools summarise faster and they can miss the specific user-agent lines that decide crawler access.
Which AI crawlers should I allow?
Allow the crawlers that fetch pages to cite them, and think separately about training-only crawlers. Google’s crawler documentation and OpenAI’s bot documentation both list the current user agents, and both are worth checking before you edit the file.
Does an AI search visibility audit guarantee citations?
No. It makes a page eligible. Whether a system cites you depends on competition, query, timing and factors nobody outside the system can see. Eligibility is the part that responds to work.
Why does my page get summarised incorrectly?
Usually because the first hundred words do not state what the page is. The system summarises what it finds first, and a page that opens with a story rather than a subject gives it nothing accurate to work with.
Is llms.txt worth adding?
It costs twenty minutes and it does no harm. Adoption is uneven, so treat it as a small addition rather than a strategy, and do not expect it to fix a page that fails the other six checks.
How often should I re-run the audit?
Quarterly for most sites, monthly for a site publishing weekly. Crawler rules drift, dates go stale, and competitors publish. Three audits give you a trend instead of a snapshot.