---
title: "Teardown: What a 40-Page AI Visibility Audit Actually Fixes"
description: "A section-by-section teardown of the typical AI visibility audit: which pages might help win citations, which only track them, and what to demand before you pay."
url: "https://groas.com/post/teardown-the-40-page-ai-visibility-audit"
image: "https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/301dd19d-1484-44a8-8177-f9cc121d5e31.png"
published: "2026-10-10T05:20:03.300Z"
modified: "2026-10-10T05:20:03.368Z"
---

October 10, 2026 · 9 min read

# Teardown: What a 40-Page AI Visibility Audit Actually Fixes

[Alexander PerelmanHead Of Product @ groas](https://groas.com/author/alexander-perelman)[LinkedIn](https://www.linkedin.com/in/alexander-433793253/)

![A glossy bound audit report duct-taped around a burst copper pipe as a patch, soaked and failing while water keeps spraying out.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/301dd19d-1484-44a8-8177-f9cc121d5e31.png)

In this article

1. [Section 1: A visibility score needs its workings](#section-1-a-visibility-score-needs-its-workings)
2. [Section 2: The prompt list tells you whose questions got measured](#section-2-the-prompt-list-tells-you-whose-questions-got-measured)
3. [Section 3: Competitor citations need URLs, not checkmarks](#section-3-competitor-citations-need-urls-not-checkmarks)
4. [Section 4: The content-gap table is where the audit gets useful](#section-4-the-content-gap-table-is-where-the-audit-gets-useful)
5. [Section 5: The technical checklist hides the urgent fixes](#section-5-the-technical-checklist-hides-the-urgent-fixes)
6. [Section 6: Recommendations expose who is doing the work](#section-6-recommendations-expose-who-is-doing-the-work)
7. [The work after the PDF: Ship, log, rerun](#the-work-after-the-pdf-ship-log-rerun)
8. [Five questions to ask before buying another audit](#five-questions-to-ask-before-buying-another-audit)

The artefact is the 40-page AI visibility audit: your logo on page one, a score on page three, and most of the remaining pages telling you where AI answers mention someone else. I chose it because businesses buy it to get featured in Google AI Overviews and cited by assistants, but the PDF usually answers a different question: _how absent are we?_

I’m walking through a composite of the standard deliverable in the order a buyer reads it: visibility score, prompt list, competitor citation table, content gaps, technical issues, recommendations. My verdict up front: **most of the audit monitors; very little of it fixes.** The two sections most likely to move citations arrive as a to-do list for a team you may not have. It is the 2015 agency audit with “AI” added to the cover.

Say you spend $20k a month and your boss forwards a ChatGPT screenshot that names a competitor but not you. The market offers free scans checking 3 to 20 prompts, [$29-to-$500-a-month tracking subscriptions](https://subscribepr.com/blog/how-much-does-an-ai-visibility-audit-cost/) like Otterly at $29 for 15 prompts or Peec around $95 for 50, and one-time audits at [$997 to $5,000, usually $2k to $3.5k for 100 pages](https://subscribepr.com/blog/how-much-does-an-ai-visibility-audit-cost/). A credible view of one market needs [40 to 120 prompts with repeats](https://subscribepr.com/blog/how-much-does-an-ai-visibility-audit-cost/). Keep that standard in mind as we turn the pages.

## Section 1: A visibility score needs its workings

Page three gives you a big number in a colored circle. 62 out of 100. Share of voice: 14%. Positive sentiment. It looks authoritative in the way Quality Score looked authoritative in 2015, right up until you ask what went into it.

A score without the prompt list, rerun count, engine split, and separate counts for mentions and linked citations tells you little. Those are the [four things to demand before trusting it](https://typescape.ai/blog/best-ai-visibility-products-geo). The problem is variance: one test ran 12 B2B prompts 100 times each and found that ChatGPT surfaced [about 44 brands across 100 runs, but only about 10 in any single answer; roughly five appeared in at least 80% of runs](https://searchengineland.com/repeated-chatgpt-runs-brand-visibility-468552). The test’s author calls single-check trackers grossly misleading and says key prompts need at least five runs.

**If the audit ran each prompt once, its 62 is a coin flip with a cover page.** Verdict: fail until it shows the repeats. Ask for the runs before you discuss the score.

## Section 2: The prompt list tells you whose questions got measured

Pages 5 to 9 are usually a neat table of 20 to 30 prompts: “best CRM for small business,” “top project management software,” “best dentist near me.” I used to tell clients a keyword list was research. Often it was inventory. A prompt list can make the same mistake.

The vendor chose these questions. [Vendor-curated sets, differences between API and consumer answers, and no access to real query streams](https://typescape.ai/blog/best-ai-visibility-products-geo) can make the resulting score hard to verify. What the deck calls coverage, I call 25 questions whose connection to your buyers has not been shown.

The fix is specific: establish a baseline on a fixed set of [25 to 50 buyer prompts and a defined competitor set](https://blog.hubspot.com/marketing/ai-search-optimization-tools). Record why each question belongs. Then [keep the prompts fixed](https://blog.hubspot.com/marketing/ai-search-optimization-tools); changing them every month destroys the comparison you paid to make.

**If the audit cannot show who chose each prompt and why a buyer would ask it, skip the score.** Verdict: useful raw material, incomplete research. Make the vendor defend the questions before you accept the answers.

## Section 3: Competitor citations need URLs, not checkmarks

Pages 10 to 14 are the grid most owners turn to first. Your brand sits beside three competitors, with checkmarks for mentions and perhaps a share number. This section points in the right direction: it shows who takes the answer when a buyer asks. But a brand name does not tell you what to fix.

Here is the test. Does the table give you the **exact cited URLs, dates, and prompts**, or only logos with checkmarks? Without the URLs, you cannot tell whether the cited answer came from a competitor’s site or somewhere else.

That distinction matters. About [79% of AI-cited sources are third-party pages rather than brand sites](https://christopherjanb.com/blog/llm-seo/). The page you need to examine may be a G2 comparison, Reddit thread, review page, or publisher listicle, not your homepage. The work that follows may involve [comparison-page insertion, review and community mentions, and consistent entity information](https://christopherjanb.com/blog/llm-seo/) across names, authors, your Google Business Profile, and Wikidata. Another rewrite of your About page will not address a missing third-party mention.

**Keep this table if it names URLs. If it only names brands, it is decoration.** Verdict: potentially useful, but only with receipts. Find the cited page before choosing the fix.

## Section 4: The content-gap table is where the audit gets useful

Pages 15 to 25 are a spreadsheet: prompt, winner, winning URL, and the angle you lack. That angle might be a pricing breakdown, comparison table, pros and cons, or a direct answer near the top of the page. This is the section I would keep.

If answers cite a competitor’s comparison page for eight buyer prompts in a row, the table identifies a gap you can close by publishing. The mechanism matters more than the score: a page that answers the buyer’s question in clear, quotable HTML gives an assistant something relevant to retrieve and cite. The table can name both the question and the page to build.

**Owning the gap table is not the same as closing the gap.** That is where the audit stalls. The market splits into [monitors and closers](https://growganic.io/blog/ai-visibility-tools-compared): Profound monitors and generates content at enterprise cost; Peec, around $95 a month, provides dashboards and share of voice but no published pages; Otterly, at about $29 a month for 15 prompts, is monitoring and audit only. Diagnosis still leaves someone to write, publish, and pursue mentions.

Suppose your $2,500 gap list calls for 12 pages, two comparison placements, and review outreach. If you have no writer, no developer hours, and a CMS nobody wants to touch, that list does not move citations. It becomes another piece of work waiting for an owner. **A gap you do not ship is monitoring with extra steps.** Verdict: keep the table, then assign the work it describes.

![A stack of printed audit pages beside a single published web page](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/a9b2e013-5709-405a-a864-bac08514ab2f.png)

## Section 5: The technical checklist hides the urgent fixes

Pages 26 to 33 look like a crawl report wearing an AI costume: broken links, missing alt text, Core Web Vitals, and a wall of green checks. Some of that work may matter to your site. It does not all deserve the same billing as an AI visibility fix. I would look first for **crawler access and readable page content**.

Start with bot access. The retrieval crawlers that fetch for answers include OAI-SearchBot, Claude-SearchBot, and PerplexityBot. Blocking GPTBot or ClaudeBot affects future training data; [blocking OAI-SearchBot removes you from ChatGPT search](https://christopherjanb.com/blog/llm-seo/). Google-Extended does not affect Search inclusion or AI Overviews. An audit that says only “check robots.txt” has not told you which access decision matters.

Next, check rendering. Tests found that major AI crawlers did not execute JavaScript. [GPTBot fetched JavaScript in 11.5% of cases and ClaudeBot in 23.84%, but neither executed it](https://blog.hubspot.com/marketing/ai-search-optimization-tools). If the answer appears only after JavaScript renders, the page source can look empty to a crawler. That belongs near the front of the audit, not buried behind a list of routine checks.

Then come the fashionable additions. Schema can help machines parse a page, but an Ahrefs controlled test of 1,885 JSON-LD pages against about 4,000 controls found [+2.4% in AI Mode and +2.2% in ChatGPT that looked like noise, and -4.6% in AI Overviews](https://blog.hubspot.com/marketing/ai-search-optimization-tools/). The case for treating _llms.txt_ as urgent is weaker still: [97% of valid files received zero requests in May 2026](https://blog.hubspot.com/marketing/ai-search-optimization-tools/), and Google says the file is not required for AI features.

**Fix fetch first, then HTML clarity; worry about markup later.** Verdict: a few checks worth acting on inside pages of less urgent work. Ask which finding prevents a crawler from reading an answer today.

## Section 6: Recommendations expose who is doing the work

Pages 34 to 40 are where the audit goes to die. Count the verbs: “Consider adding FAQ content.” “Explore strengthening entity signals.” “Ensure technical best practices are followed.” Those lines sound like direction, but none names a page, an owner, or a ship date.

The mechanism is simple. A PDF cannot publish a page, unblock a crawler, or place a comparison. Unless someone takes responsibility for those jobs, the recommendations can stay safely vague. You receive a plan that depends on your team, whether or not your team has the hours.

**If the last pages do not say who ships what by when, you bought monitoring dressed as optimization.** Verdict: flip to the back before admiring the front. Soft verbs make even a carefully measured score less useful.

## The work after the PDF: Ship, log, rerun

Getting featured is shipping, not scoring. Assistants need to fetch the material, find an answer to the buyer’s question, and encounter the sources that support it. So the useful sequence is to publish the missing pages, [remove technical blocks that keep crawlers out](https://groas.com/post/fix-technical-seo-issues-hurt-ai-visibility), pursue comparison and review mentions, then rerun the same fixed prompt set and count citations again.

Keep the questions and engines the same before and after. Log each change with a date: page shipped, block removed, mention earned. Otherwise, a new score cannot tell you much about whether the work helped. This is the distinction I come back to when assessing [tools that help you get featured rather than merely monitored](https://groas.com/post/tools-to-get-featured-in-google-ai-overviews).

**Demand execution, a change log, and remeasurement on fixed prompts.** Everything else is a weather report.

![A work log of published pages and crawler fixes beside citation measurements](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/6e892665-59c1-43ee-8ec2-c8e5483a2954.png)

## Five questions to ask before buying another audit

This is the order I would use on an audit pitch. Start with the measurement, move to the proposed fixes, and finish with proof that anyone did the work.

1. **How many times did you run each prompt, and on which engines?** I want five or more runs per key prompt and the dates. A single answer is too variable to carry the score.
2. **Can I see the prompt list and who picked it?** Show me the fixed buyer prompts and why each matches a question a customer would ask. If the rationale is missing, the score measures the vendor’s choices, not necessarily your market.
3. **Do you name cited URLs or just brands?** A checkmark beside a competitor does not identify the page that won. I want the exact sources before I decide what to publish or where to pursue a mention.
4. **Who ships the fixes, and what ships first?** Start with content gaps and crawler blocks, not _llms.txt_. If my team must do the work, tell me that plainly.
5. **How will you show a lift?** Use the same prompts and engines before and after, alongside a dated log of changes. I would press that question on any [AI visibility software claiming to win citations](https://groas.com/post/will-any-of-this-software-actually-get-m).

Overall verdict: **keep the content-gap table, the crawler and rendering checks, and the competitor table if it names URLs.** Toss the single-run score, a prompt list with no buyer rationale, the schema and _llms.txt_ padding, and every recommendation that starts with “consider.” If you have a writer, developer hours, and someone to do outreach, the gap table can be worth the PDF. If you do not, it is homework.

I used to sell versions of this audit in PPC form. I was wrong about what made them valuable. The finding was never the valuable part on its own; the value came from what got shipped afterward. Ask for the log, the ship dates, and the rerun on the same prompts. If the next seller cannot promise those three, keep your $3,000 and publish one page that answers a buyer’s question better than the page cited today.

## Frequently Asked Questions

### Why is a single-run AI visibility score unreliable?

ChatGPT answers vary heavily from run to run: one test of 12 B2B prompts run 100 times each found about 44 brands surfaced across 100 runs but only about 10 in any single answer. A score based on one run per prompt is essentially a coin flip, so key prompts need at least five runs.

### What should I check about the prompts used in an AI visibility audit?

Establish a baseline on a fixed set of 25 to 50 buyer prompts with a defined competitor set, and make the vendor explain why each question matches something a real customer would ask. Keep the prompts fixed; changing them every month destroys the comparison you paid to make.

### Why do I need the exact URLs in a competitor citation table?

About 79% of AI-cited sources are third-party pages such as G2 comparisons, Reddit threads, review pages, or publisher listicles, not brand sites. Without the exact cited URLs you cannot tell where the winning answer came from, so you cannot decide what to publish or where to pursue a mention.

### Does a content-gap table in an AI visibility audit actually win citations?

The gap table itself only diagnoses: it names the prompt, the winning URL, and the angle you lack, such as a pricing breakdown or comparison table. Citations move when you publish pages that answer the buyer's question in clear, quotable HTML, so a gap you do not ship is just monitoring with extra steps.

### Which technical fixes in an AI visibility audit actually matter most?

Crawler access and readable page content come first: blocking OAI-SearchBot removes you from ChatGPT search, while Google-Extended does not affect Search inclusion or AI Overviews. Rendering matters too, since major AI crawlers fetch JavaScript but do not execute it, so a JavaScript-only answer can look empty to them.

### Is schema markup or llms.txt urgent for AI visibility?

The evidence is weak. An Ahrefs test of 1,885 JSON-LD pages against about 4,000 controls found +2.4% in AI Mode, +2.2% in ChatGPT that looked like noise, and -4.6% in AI Overviews. Meanwhile, 97% of valid llms.txt files received zero requests in May 2026, and Google says the file is not required for AI features.

### How can I tell if an audit's recommendations will actually get done?

Look for a named page, an owner, and a ship date. A PDF cannot publish a page, unblock a crawler, or place a comparison, so recommendations that only say 'consider' or 'explore' depend entirely on your team's capacity and often stay vague.

### How do I measure whether my AI visibility work actually worked?

Publish the missing pages, remove blocks that keep crawlers out, pursue comparison and review mentions, then rerun the same fixed prompt set on the same engines and count citations again. Log each change with a date, such as a page shipped or a block removed, so the new score can tell you whether the work helped.

## Pay For Results, Not For Hours

Businesses buy the outcome, agencies resell it, and groas answers for it either way.

[See If You Qualify](https://groas.typeform.com/to/xC1bQNUT)

## Structured data

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://graph.groas.com/entity/groas","name":"groas","url":"https://groas.com/","logo":{"@type":"ImageObject","url":"https://groas.com/icon-512.png","width":512,"height":512},"sameAs":["https://www.linkedin.com/company/groas/"]},{"@type":"WebSite","@id":"https://groas.com/#website","name":"groas","url":"https://groas.com/","inLanguage":"en-US","publisher":{"@id":"https://graph.groas.com/entity/groas"}},{"@type":"WebPage","@id":"https://groas.com/post/teardown-the-40-page-ai-visibility-audit#webpage","url":"https://groas.com/post/teardown-the-40-page-ai-visibility-audit","name":"Teardown: What a 40-Page AI Visibility Audit Actually Fixes","description":"A section-by-section teardown of the typical AI visibility audit: which pages might help win citations, which only track them, and what to demand before you pay.","inLanguage":"en-US","isPartOf":{"@id":"https://groas.com/#website"},"about":{"@id":"https://graph.groas.com/entity/groas"},"primaryImageOfPage":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/301dd19d-1484-44a8-8177-f9cc121d5e31.png"},"breadcrumb":{"@id":"https://groas.com/post/teardown-the-40-page-ai-visibility-audit#breadcrumb"},"dateModified":"2026-10-10T05:20:03.368Z","mainEntity":{"@id":"https://groas.com/post/teardown-the-40-page-ai-visibility-audit#article"}},{"@type":"BreadcrumbList","@id":"https://groas.com/post/teardown-the-40-page-ai-visibility-audit#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://groas.com/"},{"@type":"ListItem","position":2,"name":"Blog","item":"https://groas.com/blog"},{"@type":"ListItem","position":3,"name":"Teardown: What a 40-Page AI Visibility Audit Actually Fixes","item":"https://groas.com/post/teardown-the-40-page-ai-visibility-audit"}]},{"@type":"BlogPosting","@id":"https://groas.com/post/teardown-the-40-page-ai-visibility-audit#article","url":"https://groas.com/post/teardown-the-40-page-ai-visibility-audit","mainEntityOfPage":{"@id":"https://groas.com/post/teardown-the-40-page-ai-visibility-audit#webpage"},"isPartOf":{"@id":"https://groas.com/blog#blog"},"headline":"Teardown: What a 40-Page AI Visibility Audit Actually Fixes","description":"A section-by-section teardown of the typical AI visibility audit: which pages might help win citations, which only track them, and what to demand before you pay.","datePublished":"2026-10-10T05:20:03.300Z","dateModified":"2026-10-10T05:20:03.368Z","image":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/301dd19d-1484-44a8-8177-f9cc121d5e31.png"},"author":{"@id":"https://groas.com/author/alexander-perelman#person"},"publisher":{"@id":"https://graph.groas.com/entity/groas"},"keywords":"audit, page, pages, visibility audit, visibility, score, prompts, fixes","wordCount":1984,"inLanguage":"en-US"},{"@type":"Person","@id":"https://groas.com/author/alexander-perelman#person","name":"Alexander Perelman","jobTitle":"Head Of Product @ groas","description":"Ex Goldman Sachs and Ex Stanford Computer Science","image":"https://groas.com/media/blog/4f17cac3a81acc224cc1b2cfab8f94bbc295a3089d13e8421125771ef8d4064f.jpg","worksFor":{"@id":"https://graph.groas.com/entity/groas"},"sameAs":["https://www.linkedin.com/in/alexander-433793253/"]},{"@type":"FAQPage","@id":"https://groas.com/post/teardown-the-40-page-ai-visibility-audit#faq","mainEntity":[{"@type":"Question","name":"Why is a single-run AI visibility score unreliable?","acceptedAnswer":{"@type":"Answer","text":"ChatGPT answers vary heavily from run to run: one test of 12 B2B prompts run 100 times each found about 44 brands surfaced across 100 runs but only about 10 in any single answer. A score based on one run per prompt is essentially a coin flip, so key prompts need at least five runs."}},{"@type":"Question","name":"What should I check about the prompts used in an AI visibility audit?","acceptedAnswer":{"@type":"Answer","text":"Establish a baseline on a fixed set of 25 to 50 buyer prompts with a defined competitor set, and make the vendor explain why each question matches something a real customer would ask. Keep the prompts fixed; changing them every month destroys the comparison you paid to make."}},{"@type":"Question","name":"Why do I need the exact URLs in a competitor citation table?","acceptedAnswer":{"@type":"Answer","text":"About 79% of AI-cited sources are third-party pages such as G2 comparisons, Reddit threads, review pages, or publisher listicles, not brand sites. Without the exact cited URLs you cannot tell where the winning answer came from, so you cannot decide what to publish or where to pursue a mention."}},{"@type":"Question","name":"Does a content-gap table in an AI visibility audit actually win citations?","acceptedAnswer":{"@type":"Answer","text":"The gap table itself only diagnoses: it names the prompt, the winning URL, and the angle you lack, such as a pricing breakdown or comparison table. Citations move when you publish pages that answer the buyer's question in clear, quotable HTML, so a gap you do not ship is just monitoring with extra steps."}},{"@type":"Question","name":"Which technical fixes in an AI visibility audit actually matter most?","acceptedAnswer":{"@type":"Answer","text":"Crawler access and readable page content come first: blocking OAI-SearchBot removes you from ChatGPT search, while Google-Extended does not affect Search inclusion or AI Overviews. Rendering matters too, since major AI crawlers fetch JavaScript but do not execute it, so a JavaScript-only answer can look empty to them."}},{"@type":"Question","name":"Is schema markup or llms.txt urgent for AI visibility?","acceptedAnswer":{"@type":"Answer","text":"The evidence is weak. An Ahrefs test of 1,885 JSON-LD pages against about 4,000 controls found +2.4% in AI Mode, +2.2% in ChatGPT that looked like noise, and -4.6% in AI Overviews. Meanwhile, 97% of valid llms.txt files received zero requests in May 2026, and Google says the file is not required for AI features."}},{"@type":"Question","name":"How can I tell if an audit's recommendations will actually get done?","acceptedAnswer":{"@type":"Answer","text":"Look for a named page, an owner, and a ship date. A PDF cannot publish a page, unblock a crawler, or place a comparison, so recommendations that only say 'consider' or 'explore' depend entirely on your team's capacity and often stay vague."}},{"@type":"Question","name":"How do I measure whether my AI visibility work actually worked?","acceptedAnswer":{"@type":"Answer","text":"Publish the missing pages, remove blocks that keep crawlers out, pursue comparison and review mentions, then rerun the same fixed prompt set on the same engines and count citations again. Log each change with a date, such as a page shipped or a block removed, so the new score can tell you whether the work helped."}}]}]}
```