Case study · anonymised · this is our published sample report

Case study: a site that passed the technical checks and still lost the AI answer

A B2B technology company had a site that was technically ready for AI search: no blocking issue, every key AI search crawler could reach it. Then Full Search Audit read what the assistants actually said. ChatGPT named the company in 0 of its 14 answers, the top competitor appeared in 39 of 41 answers against the company’s 19, and 15 answers stated a price that was not on the pricing page.

This is a real AI Search Visibility Audit, run on 24 September 2026 for the US market and anonymised for sharing. It is the same audit as our published sample report: the company is shown as Example Co., and its competitors, product category, use cases and prices are masked. Shortened sample: 8 buyer questions instead of 20, with 1 to 3 runs per question instead of 3 (41 answers).

Updated

The audit in five numbers

What did the audit find, in short?

The audit found a site that scored well on its own pages and lost on what assistants said. Five numbers from the summary page of the report:

73 / 100AI Search Readiness Score, labelled “Needs work”

19 of 41AI answers mentioned the company

0 of 14ChatGPT answers mentioned it

3 vs 8times an answer’s final pick was the company vs its top rival

15answers stated a price that did not match the company’s site

The Readiness Score is Full Search Audit’s assessment of the site. The answer counts are what ChatGPT, Perplexity, Claude and Grok said on the day. The report keeps the two apart: the answers are counted, not scored.

The situation

What did the site look like before anyone read the AI answers?

The site looked ready. The report’s summary says: “No blocking issue. Every key AI search crawler can reach the site, pages are indexable and readable without JavaScript.” On its own pages the company was ahead of the three competitors it named: the most quotable text and the richest structured data.

Readiness categoryScoreLabel
AI Crawler Access96.7Strong
Answer Format and Index Hygiene95.0Strong
Structured Data90.0Strong
Technical Foundation87.4Strong
Content Trust (E-E-A-T)63.6Needs work
AI Citability50.2At risk
Brand and Entity Authority46.7At risk

All 8 AI search crawlers checked were open, the same as for each of the three competitors. Citability of the home and pricing pages was 51.6, against 39.3, 41.5 and 39.4 for the competitors. A site check alone would have reported a healthy site with a few weaker areas.

What the assistants said

What did ChatGPT, Perplexity, Claude and Grok actually say?

Two assistants named the company often and two almost never. Perplexity named it in 7 of 8 answers and Claude in 11 of 14. ChatGPT named it in none of its 14 answers and never cited its site. Grok named it in 1 of 5.

AssistantNamed the companyCited its site
Perplexity7 of 88 of 8
Claude11 of 1413 of 14
ChatGPT0 of 140 of 14
Grok1 of 54 of 5
All four19 of 41

Across the 41 answers the company was listed first 5 times and was in the top 3 in 9 answers. Its average position was 4.2 in lists that named 6.7 providers on average.

Google AI Overviews were checked on 12 target searches. An AI Overview appeared on all 12 and cited the company’s site on 4.

Against competitors

Who did the assistants name instead?

The assistants named the top competitor in almost every answer. In the same 41 answers, Competitor A appeared in 39, Competitor B in 35, Competitor C in 24 and the audited company in 19. When an answer picked one provider, it picked Competitor A 8 times and the company 3 times.

Measured on the same questions and searchesExample Co.Competitor ACompetitor BCompetitor C
Mentioned in AI answers19 of 4139 of 4135 of 4124 of 41
Cited in Google AI Overviews4 of 123 of 125 of 120 of 12
AI search crawlers open8 of 88 of 88 of 88 of 8
Citability of home and pricing pages51.639.339.441.5

The table is the core of the case: the company had the better pages and the fewer mentions. The report’s second High finding names the reason in its title: “Competitors win on third-party proof, not on the site.”

Wrong facts

Which prices did the assistants quote?

Old ones. 15 answers stated a price that was not on the company’s pricing page. The report lists 9 different stated prices that did not match, all from Claude and Perplexity. Asked what the company is and how much it costs, Perplexity and Claude quoted outdated monthly prices, while ChatGPT’s answer to the same question matched the pricing page.

Every one of the 19 answers that mentioned the company cited its own site or blog as a source. The report’s fix for the prices: “Update or annotate older blog posts that carry old prices and link them to /pricing/.”

Why it happened

Why did a technically ready site lose the answer?

Because the gap was outside the site. The report has 13 findings: 3 High, 6 Medium and 4 Low. The three High findings explain the result.

High: ChatGPT never mentions the company

ChatGPT built its answers from competitor sites and never cited the company’s blog. The fix in the report: get listed on the third-party comparison and review pages ChatGPT cites for these questions, and make sure the key pages are indexed in Bing, which ChatGPT search partly draws on.

High: competitors win on third-party proof, not on the site

On its own pages the company led. In the answers it came fourth of four. The fix in the report: put the effort into being named on the third-party sites AI assistants cite, such as comparisons, reviews and communities, not into more on-site changes.

High: AI visibility rests on the company’s own blog

The company blog was the most cited source in Claude (13 of 14 answers), Perplexity (8 of 8) and Grok (4 of 5). Perplexity usually listed the company near the top, at an average position of 2.4. Claude, reading the same posts, put it 5th to 6th on average. The fix in the report: add independent proof that AI can cross-check, such as reviews, named customer case studies and mentions in third-party comparisons.

Medium: the blog is used, the company is not named

In 6 answers (Grok 3, Claude 2, Perplexity 1) an assistant took facts from the company’s site or blog and linked to it, but did not name the company among the providers to consider.

Medium: a weak entity footprint

No Wikidata entry listed the company’s domain as the official website, and there was no Google Knowledge Panel for the brand search. The Organization markup linked 5 social profiles, but not the review and listing pages that appeared in the brand search results.

Medium: key pages only moderately quotable

Citability averaged 50 of 100 across 11 key pages. The weakest pages split their text into short fragments and label lists. 116 of 219 sections opened with a short, self-contained answer.

The plan

What did the 30-day plan recommend?

The plan put quick technical fixes first and proof outside the site last, in four weeks grouped by type of fix. Each item carries an effort label (free fix, low or medium) and the usual owner (developer, content or marketing).

1

Week 1: technical fixes

Enable HSTS and the other missing security headers. Decide on GPTBot and Meta’s crawler, which got HTTP 429 although robots.txt allowed them: allow them in the server rate limits or block them in robots.txt. Remove the cookie banner placeholder from page content. Shorten the 11 long page titles.

2

Week 2: entity and facts

Paste the Organization JSON-LD with all official profiles. Create a Wikidata item with the official website. Update the old blog posts that quote old prices and link them to the pricing page. Add named authors with bios and Person schema to blog posts.

3

Week 3: quotable pages

Rewrite the opening of each section on four pages, the home page among them, as 40 to 60 word answers. Add question headings taken from the questions buyers ask. Add answer-first sections for the three use cases where Google AI Overviews did not cite the site.

4

Week 4: proof outside the site

Ask the third-party comparison and review pages that ChatGPT cites for these questions to include the company. Collect reviews on review sites. Check Bing Webmaster Tools indexing of the key pages.

The last item of the plan: after 30 days or more, ask the same questions again to measure the change (a re-audit, on request).

What happened after the audit?

This page does not say, because the report does not say. The sample is a single audit dated 24 September 2026: it records what assistants said that day and what the plan recommended. It contains no measurement after the fixes, and Full Search Audit does not promise mentions, citations, rankings or traffic. The honest way to see a change is the one the plan names: ask the same questions again later and compare.

What this means for you

What does this case mean for your company?

It means a clean technical report is not an answer to “does AI recommend us?”. Four things carry over from this case to any B2B site, as questions to ask rather than results to expect.

A site check and an answer check are different checks

This company had a Readiness Score of 73 and no blocking issue, and one of four assistants never named it. If you only check the site, you see the first half. That is the difference between the Readiness Audit and the Visibility Audit.

Assistants do not agree with each other

Perplexity named the company in 7 of 8 answers, ChatGPT in 0 of 14. A single chat in one assistant would have shown one of those two pictures, not both.

Your own old pages can be the source of wrong facts

The report’s fix for the outdated prices was on the company’s own blog: older posts that still carried old prices. That kind of wrong fact is in your hands to fix, once you know which pages carry it.

Being cited is not being recommended

Assistants used the company’s blog as a source in answers that did not name the company. The report counts mentions, citations and final picks separately for that reason.

Limits

How far can you generalise from one audit?

Not far, and the report says so itself. This is one company, one market and one date. The sample is shortened to 8 questions and 41 answers, while a Visibility Audit uses 20 questions, 3 runs each on 4 assistants: 240 answers. Answers were collected through the assistants’ APIs with built-in web search, which is close to, but not the same as, what a signed-in person sees in the apps. Shares are counted within the question set and against the competitors the company named: they are not a market share.

Who prepared the audit

Who prepared this audit?

Vadym Borovskyi prepared and checked it, as he does every Full Search Audit report. He has 10+ years in marketing and now builds AI automation for marketing teams. He is Top Rated on Upwork. You can read all 26 pages of the anonymised report: get the sample report.

Vadym BorovskyiPrepares and checks every report

FAQ

Questions about this case

Is this a real audit?

Yes. It is a real AI Search Visibility Audit of a B2B technology company, run on 24 September 2026 and anonymised for sharing. The company appears as Example Co., and its competitors, product category, use cases and prices are masked. It is the same audit as the sample report published on this site, shortened to 8 buyer questions and 41 answers.

Why did ChatGPT not mention the company at all?

The report does not claim to know ChatGPT’s reasons. It records what happened: in 14 answers ChatGPT built its replies from competitor sites and never cited the company’s blog, although ChatGPT’s search crawlers could reach the site. The recommended fix was to get listed on the third-party comparison and review pages ChatGPT cites, and to check Bing indexing.

What results did the company get after the audit?

The report states none, so this page states none. The sample is a single audit from one date, without a later measurement. The plan’s final step is to ask the same questions again after 30 days or more and compare. Full Search Audit offers that as an AI Re-audit on request, and never promises a result in AI answers.

Would an audit of my site find the same problems?

Not necessarily. Every finding in this case comes from this company’s pages, sources and answers. Your report is built from your own buyer questions, your competitors and what assistants say on the day. If your site turns out to have no blocking issue, the report says so, as this one did, and the findings come from the answers instead.

Which audit is this, and what does it cost?

This is the AI Search Visibility Audit: $290 (€249) excl. VAT, one-time, delivered in 5 business days after you approve the questions. Your audit is larger than the sample: 20 buyer questions on ChatGPT, Perplexity, Claude and Grok, 3 runs each, plus Google AI Overviews and up to 3 competitors. Businesses only.

What would the same audit say about your site?

The Visibility Audit answers it: 240 dated answers and you against up to 3 competitors, for $290€249, in 5 business days after you approve the questions. Or read the whole case first.

Businesses only. Prices exclude VAT. The sample report is this audit in full: PDF, 26 pages, anonymised.

Get your audit