---
title: "AI visibility audit: the first ten things we check"
url: https://marketingsignals.com/insights/ai-visibility-audit-first-ten-things-we-check
type: Article
date: 2026-11-01
author: "Gareth Hoyle"
---

# AI visibility audit: the first ten things we check

Our AI visibility audit in order: crawler access, JavaScript rendering, structured data, entity consistency, quotability, prompts and brand mentions.

## Key takeaways

- Check access first: a robots.txt line or a CDN setting that blocks AI crawlers makes every other improvement invisible.
- OpenAI uses OAI-SearchBot for ChatGPT search and GPTBot for training, so opting out of training and opting out of search are separate decisions.
- Compare raw HTML with the rendered page, because AI crawlers that do not run JavaScript may see an empty template.
- Google says no special schema or AI text files are needed for its AI features, so treat llms.txt as a small extra, not a fix.
- Size does not buy AI visibility: in our fashion index Next scored 1.6 out of 100 and Whistles, with about 1% of its traffic, scored 24.8.

Most brands that ask us about AI visibility want to know why a competitor is named in ChatGPT answers and they are not. The answer is rarely one thing. It is usually a chain: a crawler blocked by a firewall rule, content that only appears after JavaScript runs, a brand described three different ways across the web.

This is the order we work through in an audit, using our free [GEO and AI visibility checker](https://marketingsignals.com/geo-ai-visibility-checker) for the page-level checks and the method behind our [UK fashion AI visibility index](https://marketingsignals.com/llm-fashion-retailers-study) for the brand-level ones. It should help you decide what to fix first, and what to stop worrying about.

## Why AI visibility matters to the business

When a buyer asks an assistant for a shortlist, the brands it cannot read or does not trust are left off. You will not see that loss in a standard report: there is no ranking to drop and often no click to lose.

The first fixes are usually cheaper than people expect. Several of the checks below are configuration changes a developer can make in a day. The expensive mistake is spending months on content and digital PR while a robots.txt line or a CDN setting stops the work being read at all. Content, PR and technical work that AI systems cannot read earns nothing in AI answers, however good it is.

Size does not protect you either. In our UK fashion AI visibility index of 59 retailers, Next attracts an estimated 21.6 million monthly UK visits and scored 1.6 out of 100. Whistles, with about 217,000 visits, scored 24.8. Traffic did not buy a place in AI answers.

## How we run an AI visibility audit: the first ten checks

We run these in order, because each depends on the one before. There is no point improving quotability on a page a crawler cannot fetch.

1. AI crawler access in robots.txt. We list which AI user agents the site allows and separate search from training. OpenAI says [OAI-SearchBot](https://developers.openai.com/api/docs/bots) surfaces sites in [ChatGPT search](https://marketingsignals.com/service/ai-search-optimisation/chatgpt-seo) while GPTBot collects training data, and Perplexity gives the same advice for PerplexityBot, so a training opt-out and a search opt-out are separate decisions. Our checker shows a grid for eight crawlers, including Google-Extended, which Google says does not affect inclusion or ranking in Google Search.
2. The CDN and firewall. Google's [AI features guidance](https://developers.google.com/search/docs/appearance/ai-features) says crawling must be allowed by any CDN or hosting infrastructure as well as robots.txt, and Cloudflare has set new domains to block AI crawlers by default since July 2025. We check bot management settings and server logs for refused requests from AI user agents.
3. Raw HTML against the rendered page. Vercel's [analysis of AI crawler traffic](https://vercel.com/blog/the-rise-of-the-ai-crawler) in December 2024 found that none of the major AI crawlers rendered JavaScript; the ChatGPT and Claude crawlers fetched script files but did not run them. If product details or article text arrive by JavaScript, those crawlers may see an empty template. Our checker compares the two versions and warns when there is a gap.
4. Google indexing and snippet eligibility. For [AI Overviews](https://marketingsignals.com/service/ai-search-optimisation/ai-overviews) and AI Mode, Google says a page must be indexed and eligible to show with a snippet, with no extra technical requirements. We look for nosnippet, max-snippet and data-nosnippet rules left behind by old legal or template decisions, because they quietly remove pages from AI features.
5. Structured data that matches the page. We check for Organization markup on the home or about page, Article markup with a named author, Product markup where relevant, and sameAs links to official profiles. Google says no special schema is needed for its AI features, but that structured data must match the visible text. Our glossary entry on [structured data for AI search](https://marketingsignals.com/glossary/structured-data-for-ai-search) covers which types matter most.
6. Entity consistency. We compare how the brand describes itself on its about page, in its markup, on LinkedIn, review sites and directories: the name, what it does, where it operates and when it was founded. Conflicting descriptions give a model several versions of one company to choose from. We treat the about page as a priority, because assistants often draw brand descriptions from it.
7. Quotability. The checker scores each page out of 100 on a single clear H1, logical sections, lists, tables, direct-answer paragraphs, sentence length and depth. Pages that answer the question in the first sentence or two under each heading are easier to quote cleanly. The [GEO paper](https://arxiv.org/abs/2311.09735) by Aggarwal and colleagues, accepted at KDD 2024, found that adding cited sources, quotations and statistics improved visibility in generative answers by 30% to 40% on its main measure.
8. llms.txt. We check whether the site has an [llms.txt](https://marketingsignals.com/glossary/llms-txt) file and whether it is accurate, but it sits low on the list. It is a proposal first published in September 2024, and Google says you do not need AI text files or new markup to appear in its AI features. It costs little to add, and it will not fix anything above it in this list.
9. The prompt set. We write the questions a buyer would put to an assistant at each stage, from problem to category to shortlist to brand, and run them in ChatGPT, Gemini, Perplexity and Google's AI features. Because answers can change between runs, we run each prompt several times. Our guide to [choosing which prompts to track](https://marketingsignals.com/insights/how-to-choose-which-prompts-to-track-in-ai-search) explains how we pick them.
10. Mentions relative to size. Finally we measure how often the brand is mentioned and cited in AI answers against its competitors, divided by organic traffic. That is the method behind our fashion index: mention and citation counts across ChatGPT and Google AI from DataForSEO and Semrush, each divided by estimated UK traffic, scored 0 to 100 and averaged. Dividing by traffic strips out the size effect and shows who is under-represented, which is where [digital PR](https://marketingsignals.com/service/digital-pr) usually comes in.

## How to measure it and what to report upwards

We deliver the audit as a ranked fix list with an owner and an effort estimate for each item, then re-run the checks after each release. Access and rendering fixes are pass or fail, so they go to leadership as done or not done.

For progress over time we track four things: requests from AI user agents in server logs, checker scores for each page template, the share of tracked prompts where the brand is named or cited, and AI assistant sessions and key events in GA4. Search Console's generative AI performance report adds impressions in AI Overviews and AI Mode by page.

Good looks like a rising share of prompts where you are named or cited, measured against the same competitors each month. Tell the board that access fixes take effect quickly (OpenAI says robots.txt changes take around 24 hours to reach ChatGPT search), while citation share moves more slowly because it depends on what others write about you.

We also re-run the prompt set after each major fix and keep the raw answers, not just a score. When a citation appears or disappears, the saved answer shows which source the assistant used instead, which is often the most useful finding in the whole audit.

## When it does not fit, and the mistakes we see most

An audit like this matters less if almost all your demand is branded or transactional, or if you have chosen to keep content away from AI systems for licensing reasons. In those cases we confirm that access is deliberate and stop there.

- Blocking GPTBot and assuming it controls ChatGPT search. OpenAI uses OAI-SearchBot for that.
- Checking robots.txt and never looking at the CDN, which is often where the block sits.
- Testing pages in a browser, where JavaScript has already run, rather than in the raw HTML a crawler receives.
- Adding schema that says things the page does not, which Google's guidance warns against.
- Treating llms.txt as a fix rather than a small extra.
- Fixing pages one at a time when the rendering or markup problem sits in a template shared by thousands of pages.
- Running each prompt once and reporting the result as if it were a fixed position.

## Questions to ask your team or agency

This audit is the starting point of our [AI SEO and GEO work](https://marketingsignals.com/service/ai-search-optimisation). Whoever runs it, these questions show quickly whether the basics are covered.

- Which AI crawlers do we allow, and was that a decision or a default?
- Has anyone checked our CDN or firewall for blocked AI user agents?
- What does our key content look like in raw HTML, before JavaScript runs?
- Do our about page, our markup and our LinkedIn page describe the company the same way?
- Which prompts do we track, and how often are we named against our three main competitors?

## Frequently asked questions

### What is AI visibility?

AI visibility is how often, and how accurately, AI assistants such as ChatGPT, Gemini, Perplexity and Google's AI Overviews mention, cite or recommend your brand when people ask questions in your market. It depends on whether AI crawlers can read your site, whether they can trust and quote what they find, and how often credible third-party sources write about you.

### How do I check my website's AI visibility?

Start with access: check robots.txt and your CDN for blocked AI crawlers such as OAI-SearchBot and PerplexityBot. Then compare your raw HTML with the rendered page, check that structured data matches visible content and score pages for quotability. Our free Chrome extension, the GEO and AI visibility checker, runs these page checks in about 30 seconds. Then track a set of buyer prompts.

### Does blocking GPTBot remove my site from ChatGPT?

On its own, no. OpenAI uses GPTBot to collect content that may be used to train its models, and OAI-SearchBot to surface sites in ChatGPT search results. Disallowing GPTBot signals that your content should not be used for training. To appear in ChatGPT search answers, OpenAI recommends allowing OAI-SearchBot in robots.txt, and says changes take around 24 hours to apply.

## Sources

- [OpenAI: Overview of OpenAI crawlers](https://developers.openai.com/api/docs/bots)
- [OpenAI Help Center: Publishers and developers FAQ](https://help.openai.com/en/articles/12627856-publishers-and-developers-faq)
- [Perplexity: Perplexity crawlers](https://docs.perplexity.ai/guides/bots)
- [Google Search Central: AI features and your website](https://developers.google.com/search/docs/appearance/ai-features)
- [Google crawling documentation: Google's common crawlers (Google-Extended)](https://developers.google.com/crawling/docs/crawlers-fetchers/google-common-crawlers)
- [Cloudflare press release: Cloudflare just changed how AI crawlers scrape the internet at large (1 July 2025)](https://www.cloudflare.com/press/press-releases/2025/cloudflare-just-changed-how-ai-crawlers-scrape-the-internet-at-large/)
- [Vercel: The rise of the AI crawler (December 2024)](https://vercel.com/blog/the-rise-of-the-ai-crawler)
- [Google Search Central: Organization structured data](https://developers.google.com/search/docs/appearance/structured-data/organization)
- [llmstxt.org: The /llms.txt file proposal](https://llmstxt.org/)
- [Aggarwal et al., GEO: Generative Engine Optimization (arXiv 2311.09735, KDD 2024)](https://arxiv.org/abs/2311.09735)
- [Google Search Central Blog: Introducing Search Generative AI performance reports in Search Console](https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports)
- [Marketing Signals: Top UK Fashion and Apparel AI Visibility Index](https://marketingsignals.com/llm-fashion-retailers-study)
- [Marketing Signals: GEO and AI Visibility Checker](https://marketingsignals.com/geo-ai-visibility-checker)

Visibility Playbooks

## Related Playbooks

More Insights

## More articles we think you’ll like