---
title: "Teardown: The 42-Page AI Visibility Audit With Six Useful Findings"
description: "A finding-by-finding teardown of a $3,500 GEO audit: which six items point to citation barriers, and why the other 41 make for a costly to-do list."
url: "https://groas.com/post/teardown-the-40-page-ai-visibility-audit-2"
image: "https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/48149e2c-60a1-4084-ab60-870d56a1d41d.png"
published: "2026-10-14T05:53:00.980Z"
modified: "2026-10-14T05:53:01.049Z"
---

October 14, 2026 · 9 min read

# Teardown: The 42-Page AI Visibility Audit With Six Useful Findings

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

![A thick audit report hollowed out with a craft knife, paper scraps around it, with only six small red-tabbed slips kept standing on top.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/48149e2c-60a1-4084-ab60-870d56a1d41d.png)

In this article

1. [Page two: A visibility score with no useful next move](#page-two-a-visibility-score-with-no-useful-next-move)
2. [Pages 4–19: Eighteen technical findings, two worth pursuing](#pages-419-eighteen-technical-findings-two-worth-pursuing)
3. [Pages 20–31: The content gaps hide three buyer questions](#pages-2031-the-content-gaps-hide-three-buyer-questions)
4. [Pages 32–37: The competitor chart conceals its best finding](#pages-3237-the-competitor-chart-conceals-its-best-finding)
5. [Pages 38–42: An action plan that hands the work back](#pages-3842-an-action-plan-that-hands-the-work-back)
6. [Overall verdict: Keep the six findings tied to an answer](#overall-verdict-keep-the-six-findings-tied-to-an-answer)

A 42-page AI visibility audit can look like two weeks of urgent engineering work. This one carries a $3,500 invoice, a **41-out-of-100 score**, 47 findings, and an estimate of 95 developer hours to clear them. I chose it for this teardown because it poses the question behind every AEO or GEO content-gap audit: which of those tickets would change whether an AI engine cites the business?

My answer is six. The other 41 range from ordinary site maintenance to speculative tasks and recommendations too vague to execute. That does not make every one of them bad for a website. It makes them a poor answer to a question about citations. I am going through the PDF in the order its reader encounters it: the score, the technical warnings, the content gaps, the competitor charts, and the action plan. The test at each stop is the one I used to apply to agency handover decks: **what would actually move the result?**

## Page two: A visibility score with no useful next move

The first substantial page presents a dial reading 41 out of 100, flanked by sub-scores for “AI Authority,” “Crawler Readiness,” and “Semantic Density.” Audits in this category, whether produced by a platform or an agency’s white-label script, use composite numbers to establish urgency ([one overview of GEO audit tools](https://specularis.io/blog/best-free-ai-visibility-geo-aeo-audit-tools-2026/) describes the pattern). The methodology appendix is less imposing: 30% traditional crawl warnings, 30% presence of speculative metadata files, and 40% raw brand mentions across 25 unweighted prompts.

That is not a citation benchmark. It is a rubric for the things the auditor chose to count. A [Growth and Convert case study](https://www.growthandconvert.com/how-to-run-an-ai-visibility-audit/) shows why the distinction matters: a brand recorded a strong AI Visibility rating and 9,900 brand mentions, yet had under half the actual citations and cited source pages of its main competitor. The mentions did not tell the brand which pages answered buyers’ questions. The competitor’s comparison guides did.

The score also creates a predictable incentive. Give a team a target of 75 and it can improve the number by clearing the cheapest rubric items: an alt-text pass here, a longer page there. The difficult question—what specific buyer question does the site fail to answer?—stays untouched.

![A conceptual risograph print showing a thick audit report outweighed on a brass scale by a single index card.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/f555543b-cabf-4e13-b6b2-302770390d46.png)

**Verdict: Cut the composite score.** Keep the underlying observations if they identify a page, a prompt, and a plausible citation barrier. Otherwise, page two tells you how to please the audit, not what to fix.

## Pages 4–19: Eighteen technical findings, two worth pursuing

The technical section opens with three “Critical Emergencies”: no `llms.txt` file, incomplete JSON-LD Organization schema on 42 URLs, and a mobile Largest Contentful Paint (LCP) of 3.1 seconds. I would not call site speed or structured data worthless. I would call it a mistake to sell these three items as the urgent explanation for missing AI citations.

The numbers cited in the draft’s supporting material undercut that pitch. [Search Engine Journal’s analysis](https://searchenginejournal.com/9-ai-search-myths-debunked-by-15-million-data-points/) reports that 97% of valid `llms.txt` files in its dataset received no crawler requests. Its schema test found a +2.2% change in ChatGPT citations after adding JSON-LD to 1,885 pages, a result it did not distinguish statistically from zero. [Tracking of AI-bot requests](https://theseoframework.com/blog/ai-seo-myths-and-facts/) points toward extraction of page content rather than a dependable citation benefit from metadata alone. None of that says to neglect technical SEO. It says not to spend a citation-focused sprint on metadata simply because the audit labels it critical.

Two findings do describe a direct obstruction:

1. **Finding 1: Live retrieval blocked in `robots.txt`.** A business can block the wrong crawler while trying to restrict AI training. [AI Crawler Check](https://aicrawlercheck.com/robots-txt-best-practices-ai-seo-2026/) distinguishes `GPTBot`, used for training, from `OAI-SearchBot`, used for search retrieval. Blocking `Google-Extended` likewise is not the same as blocking the standard `Googlebot` used by Google AI Overviews. The task is to check the rules against the crawlers you intend to allow.
2. **Finding 4: Answers trapped behind client-side JavaScript.** [Reporting on Vercel and MERJ’s crawler research](https://www.searchenginejournal.com/technical-seo-for-ai-citations-which-claims-hold-up/) describes AI crawlers fetching raw HTML without executing client-side JavaScript. If the raw HTML omits the feature matrix or pricing table and those details appear only after rendering, the crawler may miss the facts a buyer asked for.

![A cutaway diagram showing an AI web crawler reading raw HTML while client-side JavaScript sits outside its view.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/11b01a93-8c7b-450f-9d9b-69617ef46302.png)

The remaining 16 findings include stylesheet minification, PNG compression, and meta-description rewrites. Those may belong in a normal maintenance queue. The audit has not shown why they belong in a 95-hour AI-citation rescue plan.

**Verdict: Keep two findings; remove 16 from this plan.** First confirm that retrieval crawlers can reach the relevant pages and that those pages serve the answers in accessible HTML. Give ordinary site hygiene its own justification and priority.

## Pages 20–31: The content gaps hide three buyer questions

The PDF then lists 16 “Content Gaps.” Its method looks familiar if you have ever read a conventional SEO audit: find a topic a competitor covers, call its absence a gap, and recommend a longer page. Here that produces a proposed 2,500-word guide to supply-chain digital transformation and six top-of-funnel glossaries.

I used to treat a missing category keyword as a fairly good content assignment. It is not a sufficient test here. As [Omnia’s discussion of AI visibility monitoring](https://omnia.so/blog/how-to-monitor-ai-search-visibility-2026/) argues, the useful unit is often a constrained question, not a broad keyword. Consider: _Which B2B billing software supports multi-currency invoicing under $1,000 a month?_ A long article about billing trends cannot supply a pricing threshold or confirm a feature. A clear product specification can.

This is where the audit briefly gets useful. Of its 16 content recommendations, three identify facts a buyer might reasonably need but cannot readily find on the site:

- **A transparent pricing-tier table** that lets a reader check a budget constraint.
- **A page documenting webhook capability** rather than leaving an integration question unanswered.
- **Comparative feature matrices** against the legacy market leader, so a buyer can examine the difference that matters to a switch.

Those are not glamorous assignments. They are answerable assignments. If the facts are available and accurate, publishing them gives an AI system something specific to retrieve and cite. It also gives a human buyer something more useful than another glossary entry.

![A split view comparing a buzzword-heavy marketing page with a clear table of pricing and technical specifications.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/32852714-6bab-4957-bc37-1c048dba807b.png)

The audit misses another part of the same problem. In [Search Engine Journal’s analysis of 750 commercial recommendation prompts](https://searchenginejournal.com/9-ai-search-myths-debunked-by-15-million-data-points/), third-party listicles, review directories, and independent comparison roundups accounted for 43.8% of cited URLs. An internal content-gap list cannot, by itself, tell you what those outside sources say or whether they include you. Fifteen self-congratulatory posts on your own blog will not substitute for credible external coverage.

**Verdict: Keep three findings; cut 13 generic article assignments.** Work from missing buyer facts, not a target word count. Then check the outside pages the engines cite, rather than pretending every gap lives on your domain.

## Pages 32–37: The competitor chart conceals its best finding

Next come horizontal bars comparing the client with three competitors across 25 prompts. Competitor A appears in 64% of responses; Competitor B in 48%; the client in 12%. Findings 35 through 41 frame that difference as a “brand semantic authority” crisis.

Before assigning a crisis, I want to see the prompts. Broad category questions such as _best project management software_ favour established names with extensive historical coverage. Compare a $5M challenger with a $500M incumbent on prompts that ignore why someone might switch, and the bar chart will mostly tell you which company is already larger. That can be worth knowing. It is not yet a work order.

The cited URLs offer a better lead. The audit identifies eight side-by-side migration pages from Competitor A. They address concrete switching questions: data export formats, API mapping, and per-seat cost differences. That is a mechanism I can inspect. A buyer asks how to move from a legacy tool; a page lays out the steps and constraints; the engine has a relevant source to retrieve. **Finding 37, which points to those migration tables, is the actionable finding in this section.**

The other six findings ask for broader category mindshare or an increased off-page brand footprint. Neither instruction says which buyer question is unanswered, where the current answer sits, or what the team should produce.

**Verdict: Keep one finding; cut six.** Do not try to win an unweighted prompt chart by chasing an abstract authority score. Examine the competitor’s cited pages and the questions those pages answer.

## Pages 38–42: An action plan that hands the work back

The closing “Strategic Action Plan” contains Findings 42 through 47. Its opening verbs do most of the explaining: _“Consider”_ a topical authority hub, _“Explore”_ podcast sponsorships for brand entity validation, _“Evaluate”_ semantic keyword density, and _“Monitor”_ weekly citations in a dashboard.

I mark up handover decks by circling verbs. _Consider_ is not an assigned task. _Explore_ does not specify what evidence would make a sponsorship worthwhile. _Monitor_ can tell you a problem persists, but it does not resolve it. For a PDF that estimates 95 developer hours, this is an odd final handoff: the client pays for the diagnosis and then gets the labour back.

That is the distinction I made in an earlier piece on whether [AI visibility software gets you cited or merely tracks who did](https://groas.com/post/will-any-of-this-software-actually-get-m). Measurement matters when it points to an action. Without a named page, a missing fact, and a way to check the change, the dashboard becomes a tidy display of unfinished work.

**Verdict: Cut all six closing findings.** None turns the audit’s useful observations into a publishable page or a specific technical change. They are suggestions, not the fix attached to the gap.

## Overall verdict: Keep the six findings tied to an answer

The count is less complicated than the PDF makes it look: **two technical access findings, three missing-fact findings, and one competitor-page finding.** That is six of 47. The remaining 41 may include routine upkeep, but the audit does not connect them to the citation outcome it was hired to improve.

| Where the finding sits | Keep                                                                                   | Leave out of the AI-citation plan                                                               |
| :--------------------- | :------------------------------------------------------------------------------------- | :---------------------------------------------------------------------------------------------- |
| Technical access       | Check live retrieval rules and make core answers available in the HTML crawlers fetch. | Treat `llms.txt`, schema warnings, and routine performance tasks as proof of a citation crisis. |
| Factual content        | Publish accurate pricing tiers, webhook documentation, and comparative feature facts.  | Commission broad glossaries and long guides solely because a topic is missing.                  |
| Competitor benchmark   | Examine the cited migration tables and the switching questions they answer.            | Assign work from unweighted mention charts and calls for more “mindshare.”                      |
| Closing plan           | Turn retained findings into specific work with an observable result.                   | Pay for another round of “consider,” “explore,” and “monitor.”                                  |

![A modernist graphic poster contrasting the number 6 in white on dark navy with the number 41 in faded orange.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/2107b03e-d5c5-4aff-a44b-3edf27f3e16d.png)

This is also where I would separate a gap report from an operating model. As I argued when comparing [tools aimed at getting a business featured in Google AI Overviews with tools that only monitor it](https://groas.com/post/tools-to-get-featured-in-google-ai-overviews), a list of deficits is useful only if someone takes responsibility for closing them. At [groas](https://groas.com/), our approach joins paid and organic search execution with a named human strategist who sets direction and guardrails. Specialized models work continuously, actions are logged, and the fee is flat rather than a percentage of spend. That is the alternative to treating the PDF itself as the deliverable.

My overall verdict on this audit is simple: **keep the six findings that expose a blocked answer or a missing one.** I would take the crawler-access checks, the three buyer-facing fact gaps, and the competitor’s migration-page example into a working plan. I would not send an engineering team through 41 other tickets in the hope that clearing the list makes a citation appear. A well-formatted to-do list is still a to-do list.

## Frequently Asked Questions

### Does a high AI visibility score mean an AI engine will cite my website?

Not necessarily. A composite score is a rubric for the items the auditor chose to count, such as crawl warnings, metadata files, and raw brand mentions. A brand can score well on AI visibility while its competitor earns more actual citations, so the score alone does not show which pages answer buyers' questions.

### Should I rush to add llms.txt and JSON-LD schema to get more AI citations?

Probably not as an urgent fix. One dataset found 97% of valid llms.txt files received no crawler requests, and adding JSON-LD to 1,885 pages showed a +2.2% change in ChatGPT citations that was not distinguished statistically from zero. These tasks may still be worth doing, but they are not a proven citation barrier.

### Can robots.txt block my site from appearing in AI search results?

Yes, if you block the wrong crawler. GPTBot is used for AI training while OAI-SearchBot handles search retrieval, and blocking Google-Extended is not the same as blocking the Googlebot used by Google AI Overviews. Check your robots.txt rules against the crawlers you intend to allow.

### Do AI crawlers see content that only appears after JavaScript renders my page?

No. AI crawlers generally fetch the raw HTML without executing client-side JavaScript. If a pricing table or feature matrix only appears after rendering, the crawler may miss the facts a buyer asked for, so core answers should be present in the raw HTML itself.

### What kind of content actually helps an AI engine cite a business?

Specific, answerable facts a buyer needs: a transparent pricing-tier table, documentation of capabilities such as webhooks, and comparative feature matrices against the main competitor. A clear product specification lets an AI system retrieve and cite something concrete, while broad glossaries and long trend guides rarely do.

### Do AI engines cite outside sources rather than a company's own blog posts?

Often yes. In one analysis of 750 commercial recommendation prompts, third-party listicles, review directories, and independent comparison roundups accounted for 43.8% of cited URLs. An internal content-gap list cannot show what those outside sources say, so credible external coverage matters alongside on-site content.

### Are recommendations like 'consider a topical authority hub' or 'explore podcast sponsorships' useful audit findings?

Not on their own. Verbs like consider, explore, evaluate, and monitor describe suggestions rather than assigned tasks, and monitoring can only confirm a problem persists. A useful finding names a page, a missing fact, and a way to check the change, otherwise the client pays for the diagnosis and gets the labour back.

## 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-2#webpage","url":"https://groas.com/post/teardown-the-40-page-ai-visibility-audit-2","name":"Teardown: The 42-Page AI Visibility Audit With Six Useful Findings","description":"A finding-by-finding teardown of a $3,500 GEO audit: which six items point to citation barriers, and why the other 41 make for a costly to-do list.","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/48149e2c-60a1-4084-ab60-870d56a1d41d.png"},"breadcrumb":{"@id":"https://groas.com/post/teardown-the-40-page-ai-visibility-audit-2#breadcrumb"},"dateModified":"2026-10-14T05:53:01.049Z","mainEntity":{"@id":"https://groas.com/post/teardown-the-40-page-ai-visibility-audit-2#article"}},{"@type":"BreadcrumbList","@id":"https://groas.com/post/teardown-the-40-page-ai-visibility-audit-2#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: The 42-Page AI Visibility Audit With Six Useful Findings","item":"https://groas.com/post/teardown-the-40-page-ai-visibility-audit-2"}]},{"@type":"BlogPosting","@id":"https://groas.com/post/teardown-the-40-page-ai-visibility-audit-2#article","url":"https://groas.com/post/teardown-the-40-page-ai-visibility-audit-2","mainEntityOfPage":{"@id":"https://groas.com/post/teardown-the-40-page-ai-visibility-audit-2#webpage"},"isPartOf":{"@id":"https://groas.com/blog#blog"},"headline":"Teardown: The 42-Page AI Visibility Audit With Six Useful Findings","description":"A finding-by-finding teardown of a $3,500 GEO audit: which six items point to citation barriers, and why the other 41 make for a costly to-do list.","datePublished":"2026-10-14T05:53:00.980Z","dateModified":"2026-10-14T05:53:01.049Z","image":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/48149e2c-60a1-4084-ab60-870d56a1d41d.png"},"author":{"@id":"https://groas.com/author/alexander-perelman#person"},"publisher":{"@id":"https://graph.groas.com/entity/groas"},"keywords":"audit, findings, six, pages, visibility, useful, teardown, page","wordCount":1923,"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-2#faq","mainEntity":[{"@type":"Question","name":"Does a high AI visibility score mean an AI engine will cite my website?","acceptedAnswer":{"@type":"Answer","text":"Not necessarily. A composite score is a rubric for the items the auditor chose to count, such as crawl warnings, metadata files, and raw brand mentions. A brand can score well on AI visibility while its competitor earns more actual citations, so the score alone does not show which pages answer buyers' questions."}},{"@type":"Question","name":"Should I rush to add llms.txt and JSON-LD schema to get more AI citations?","acceptedAnswer":{"@type":"Answer","text":"Probably not as an urgent fix. One dataset found 97% of valid llms.txt files received no crawler requests, and adding JSON-LD to 1,885 pages showed a +2.2% change in ChatGPT citations that was not distinguished statistically from zero. These tasks may still be worth doing, but they are not a proven citation barrier."}},{"@type":"Question","name":"Can robots.txt block my site from appearing in AI search results?","acceptedAnswer":{"@type":"Answer","text":"Yes, if you block the wrong crawler. GPTBot is used for AI training while OAI-SearchBot handles search retrieval, and blocking Google-Extended is not the same as blocking the Googlebot used by Google AI Overviews. Check your robots.txt rules against the crawlers you intend to allow."}},{"@type":"Question","name":"Do AI crawlers see content that only appears after JavaScript renders my page?","acceptedAnswer":{"@type":"Answer","text":"No. AI crawlers generally fetch the raw HTML without executing client-side JavaScript. If a pricing table or feature matrix only appears after rendering, the crawler may miss the facts a buyer asked for, so core answers should be present in the raw HTML itself."}},{"@type":"Question","name":"What kind of content actually helps an AI engine cite a business?","acceptedAnswer":{"@type":"Answer","text":"Specific, answerable facts a buyer needs: a transparent pricing-tier table, documentation of capabilities such as webhooks, and comparative feature matrices against the main competitor. A clear product specification lets an AI system retrieve and cite something concrete, while broad glossaries and long trend guides rarely do."}},{"@type":"Question","name":"Do AI engines cite outside sources rather than a company's own blog posts?","acceptedAnswer":{"@type":"Answer","text":"Often yes. In one analysis of 750 commercial recommendation prompts, third-party listicles, review directories, and independent comparison roundups accounted for 43.8% of cited URLs. An internal content-gap list cannot show what those outside sources say, so credible external coverage matters alongside on-site content."}},{"@type":"Question","name":"Are recommendations like 'consider a topical authority hub' or 'explore podcast sponsorships' useful audit findings?","acceptedAnswer":{"@type":"Answer","text":"Not on their own. Verbs like consider, explore, evaluate, and monitor describe suggestions rather than assigned tasks, and monitoring can only confirm a problem persists. A useful finding names a page, a missing fact, and a way to check the change, otherwise the client pays for the diagnosis and gets the labour back."}}]}]}
```