In plain words
This firm’s website is fast and well written, but search engines cannot reach most of its articles, and when an AI assistant recommends the firm it sends readers to a competitor’s site. The audit shows why, and what to change.
From a real audit, names removed · A professional-services firm that publishes its own research · audited in Sep 2026
- 80 of 87
- Articles a crawler cannot reach
- None of ten
- The firm’s own site among the AI answer’s sources
- None of 87
- Research pieces with an author’s name
- 95
- Lighthouse mobile, home page
Words this audit uses
- AI answer engine:
- A tool such as ChatGPT or Perplexity that answers a question in sentences and names the websites it drew on.
- Crawler:
- The program a search engine uses to read the pages of a site.
- Sitemap:
- The list of pages a site hands to search engines.
- H1, main heading:
- The main heading of a page: the line that tells a search engine what the page is about.
- Canonical tag:
- A line in a page’s code that names the one official address of that page.
- Structured data, schema, JSON-LD:
- Labels in a page’s code that tell machines what something is: an article, a product, a business.
- Redirect:
- An instruction that sends visitors and search engines from one address to the right one.
- robots.txt:
- A small file that tells crawlers which parts of a site to stay out of.
- Lighthouse:
- Google’s free test of how fast a page loads.
- LCP:
- Largest Contentful Paint: how long it takes for the main content of a page to show.
- Meta description:
- The short summary shown under a page’s title in search results.
- Alt text:
- A written description of an image, read by search engines and screen readers.
- Search Console:
- Google’s free tool that shows how a site appears in its search results.
The audit’s short version
A fast site with hand-written metadata that already ranks first or third on four commercial searches. Search engines cannot reach 80 of its 87 articles, nothing on the site carries an author’s name, and when an AI assistant recommends the firm it credits a competitor’s website as the source.
How it was measured
All 141 sitemap URLs crawled plus every internal link found on them, Lighthouse 12 run against the home page and an article, search positions checked by hand, logged out, first twenty results, and a buying question put to an AI answer engine.
What is already working
- Speed: Lighthouse mobile 95 on the home page and 98 on an article. LCP 2.7s, CLS 0.045. Nobody should sell this site a speed package.
- Metadata, across all 141 URLs: no missing meta descriptions, no duplicate titles, no missing H1s, no images without alt text, no canonical mismatches.
- Redirects: http, www and trailing-slash URLs all 301 correctly to the canonical address. A missing page returns a real 404, not a soft 200.
- The research: 87 articles, median 2,034 words, none thin, several built on the firm’s own data.
What the audit found
1AI assistants name the firm, then credit a competitor
We asked Perplexity which firms a company from abroad should hire for the firm’s main service. The firm appears in the answer, as one of the mid-market options, and the source the engine credits for it is a competitor’s website. None of its ten sources was the firm’s own site. Every other firm named got a fuller description in its own words. A second test explains why: page one of the search results for the “top firms” query in its field is eight third-party round-ups, and the firm appears in none of them. Language models do not rank vendor websites for “who should I hire” questions. They read the round-ups and repeat what those say.
Buying question put to the AI engine
Result: Firm named, a competitor’s site credited
Sources the answer cited
Result: Ten, none of them the firm’s own site
Search page one for the “top firms” query
Result: Eight round-ups, the firm in none
Test Result Buying question put to the AI engine Firm named, a competitor’s site credited Sources the answer cited Ten, none of them the firm’s own site Search page one for the “top firms” query Eight round-ups, the firm in none 2The expertise is invisible to machines
Every page on the site emits the same single block of structured data: Organization, ContactPoint, PostalAddress. Exactly one page adds anything more. Across the whole domain there is no Article markup on 87 research pieces, no BreadcrumbList, no Person, and no Service markup on 40 service pages. The breadcrumb component exists, but the stylesheet collapses it to nothing and hides it outright on mobile. Not one of the 87 research pieces carries an author’s name.
Fix: Ready-to-paste JSON-LD for Article, BreadcrumbList, Service, Person and LocalBusiness, written against the site’s actual templates. Breadcrumbs un-hidden, the author placeholder removed, heading order fixed.
3Search engines cannot walk into the archive
The research index renders seven article links in its HTML. The other eighty arrive only when a visitor clicks Load More, which fires a request to an endpoint the site’s own robots.txt tells crawlers not to visit. There is no path from any page on the site into 80 of the 87 articles. Lighthouse confirms it independently: the crawlable-anchors audit fails on the article template. Measured from the full internal link graph, 21 pages have zero internal links pointing at them and 27 have one or fewer.
Fix: A written specification for crawlable, paginated links across the archive, so all 87 articles sit behind a followable link.
4Where the site ranks today
Checked by hand, logged out, first twenty results. No rank tool and no estimates. The service that misses the top 20 has a 2,067-word page and no visibility at all.
[service A] firm [country]
Page: Service page
Position: 1
[service B] firm [country]
Page: Service page
Position: 1
[service C] firm [country]
Page: Service page
Position: 2
[the firm’s field] firm [country]
Page: Home page
Position: 3
[service D] firm [country]
Page: Service page
Position: Not in top 20
top [field] firms in [country]
Page: None
Position: Not in top 20
Search Page Position [service A] firm [country] Service page 1 [service B] firm [country] Service page 1 [service C] firm [country] Service page 2 [the firm’s field] firm [country] Home page 3 [service D] firm [country] Service page Not in top 20 top [field] firms in [country] None Not in top 20 5The remediation list
24 broken internal links with their source pages, 11 duplicate URLs with redirect targets, and 21 articles missing from the sitemap.
The order of work
- Open the archive first: there is no point building authority for pages a crawler cannot reach.
- Add the structured data and the author names.
- Clear the remediation list: broken links, duplicate URLs, the sitemap.
- Record a Search Console and analytics baseline, so later months are compared with month zero.
What this audit did not check
- No backlink counts and no keyword volumes appear in this audit. The tool pulls had not been run, and a number is left out before it is estimated.
- A number of AI citations cannot be promised by anyone: nobody controls what a language model says.
What was removed, and what was not. The business, its website and page addresses, people, competitors, places and the search terms that would name it are removed or replaced with a word in [brackets]. The prices and the proposal that closed the audit are left out. Every figure is the audit’s own.
Your own report. This audit put one buying question to one AI engine. The AI Visibility Audit runs a set of prompts across ChatGPT, Perplexity and Google’s AI Overviews, and quotes each prompt and answer in full where this sample paraphrases them to keep the firm unnamed.