New in Answer Checkup reports: listing searches, Wikidata checks and real-browser rendering
When a buyer asks ChatGPT or Perplexity for a tool recommendation, both can search the web before they answer. Whether they name you depends partly on what the web says about you, such as listings and Wikidata, and partly on whether your own pages are readable. The two paid reports, Lite and Full, now check more of both. This post covers what each new check does, what its result looks like and what it leaves out. It does not promise a ranking. It tells you what the report found.
How the answers are measured
You get a clear idea of what sits behind each answer. We measure through the engines' APIs, not screenshots: these are runs with web search, not the consumer ChatGPT and Perplexity apps. The report shows the date each engine was run, and no model details.
Lite asks each question once per engine. One run is a snapshot, not an average. Full asks each question 3 times per engine on the same day and pools every answer, with no majority vote. That is still a reading on one day. For the question side, see Does ChatGPT recommend my business?
New in both reports: third-party listing searches
You see whether the main software profile sites turn up in search for your business. For a software site, the report runs one web search per site for G2, Capterra, Crunchbase and LinkedIn. A link counts as found only when it is on that site and its address names your business. Each site gets one line:
<site>: listing found (<link>)<site>: not found in search<site>: not checked (search failed), used when the search errored or timed out
The report did not visit those sites directly, so a page the search missed may still exist. That is why the wording stays at "not found in search" and goes no further. If you see it, look at the site yourself before you draw a conclusion.
The next-steps section (in Full, part of the 30-day plan) also now points to these listing results where a profile step would otherwise say listings were not checked.
New in both reports: the Wikidata official-website check
You learn whether Wikidata ties your brand name to your domain. Wikidata is the free knowledge base many AI systems and search engines draw on. The report looks up Wikidata items for your brand name and compares the official website each item lists with the domain it scanned. Prefixes such as www are ignored. On shared hosts such as github.io, the full address must match.
There are four possible results:
- Confirmed: the item lists your domain as its official website.
- "likely, not confirmed": the item is on the same shared host but at a different address, or it has the same name but lists no official website.
- "different entity or outdated website": an item with the same name lists a different official website.
- No item matches: no item has your brand name, and none of the search hits lists your domain.
On a shared host such as github.io, the result is "confirmed" only when the item lists your exact address. An item at another address on the same host reads "likely, not confirmed". This check is one of the signals discussed in AI entity signals.
New in the Full report: rendering, crawl sample and content check
Real-browser rendering
You see what a crawler that does not run JavaScript might miss. Full loads the homepage in a real browser when browser time is available, and the report tells you if it could not. It also loads up to 4 more crawled pages whose raw HTML was thin or blocked. For each page it shows the word count in the raw HTML next to the word count after JavaScript ran.
Most AI crawlers do not run JavaScript, so text that only appears after JavaScript runs may be missed by them. Pages that robots.txt asks the crawler to skip are never loaded.
Browser rendering has a monthly allowance of 8 browser hours, shared across all reports. When the allowance is used up, the report says so in plain words. For background, see JavaScript and AI crawlers and robots.txt for AI crawlers.
Crawl sample
You find out which pages the technical crawl looked at. It covers up to 75 pages on your site. The report now says how the sample was picked: the homepage, pages picked evenly from your sitemap, pages found by following links, and pages found in links on the rendered homepage. It also says how many pages robots.txt asked it to skip. For a wider view of crawler access, see Can AI crawlers see my site?
Content check
You get a read on whether your key pages explain your business. The check covers up to 3 pages: the homepage and the 2 crawled pages with the most text. Each page gets 0 to 2 points on four things:
- what the business does
- specific facts
- answers to questions
- kept up to date
The last one follows a fixed rule on dates shown on the page: 2 for a date within 12 months of the run, 1 for an older date, 0 for no date. The other three are scored by an AI step that reads the page text against a fixed rubric, with no web search. Treat the result as a guide to what a reader finds on the page. It is not a fact check and not a promise about results.
In the 30-day plan, the actions come from a fixed playbook tied to findings in the report. An AI writing step (no web search) puts them in order and writes the one-line reasons. It also picks and words the on-page gap notes from candidates our rules found. If its text has a number we can't find in the evidence, we use our rule wording instead (small counts such as 1 to 4 are allowed).
What the reports do not check
A report that claims to cover everything is hard to trust, so here is what is left out:
- Search Console data: it needs the site owner's own Search Console access, which an outside scan can't see.
- Backlinks: not included in this report.
- Page speed: not checked.
- Whether the facts on your pages are right: the content check scores what the page text covers, not its accuracy.
Which report has what
Both reports already covered share of voice in ChatGPT and Perplexity answers against up to 5 competitors you choose, the sites the answers cite, how AI describes your brand, and short next steps (in Full, part of the 30-day plan). On top of that:
- Lite: one run per question per engine, the listing searches and the Wikidata check.
- Full: everything in Lite, plus 3 runs per question per engine (pooled), the technical crawl with AI crawler rules, real-browser rendering when browser time is available, the crawl sample note, the content check, and a ranked 30-day plan with on-page gap notes.
Prices are on the pricing page. Monthly tracking is unchanged, and the new checks are not part of it.
Answer Checkup is built largely by AI agents, with one person directing. To see your own results, start here: Run the free scan