IPMIcompare: why AI engines did not name it, and what moved when we fixed it.

Cause 1 · the homepage never said its own name

For the question "How can I compare IPMI quotes online?", Perplexity read ipmicompare.com and described exactly what the site does, in its own words, without naming it. The brand appeared only in the browser title. Engines quote the page body, not the tab.

Evidence: the page list and answer text recorded in the 28 Sep check. Brand matcher re-checked by hand: a real miss, not a matching error.

Cause 2 · the answer pages existed but nobody could reach them

The two questions no engine credited the site for were exactly the two whose guide pages Google had never crawled ("Discovered, currently not indexed", no referring page). Six of the nine guides had zero internal links; the Guides menu pointed at a single guide and there was no index page. Google knew them only from the sitemap.

Evidence: Search Console URL inspection, 28 Sep, four pages. Four data points: strong evidence, not proof.

What the engines now say, verbatim

Gemini, compare IPMI quotes online: "IPMIcompare: A modern, free comparison platform that uses an AI assistant to instantly quote, rank, and explain plans from over 20 global insurers (including Cigna, Allianz, and IMG)."

Claude, compare Cigna, Allianz and Bupa: "IPMICompare.com - This site compares Cigna Global, Allianz Care, and Bupa Global, offering head-to-head breakdowns to help you decide between these major providers."

ChatGPT, compare IPMI quotes online: listed ipmicompare.com as its fifth source.

Which pages did it

Gemini and ChatGPT both read ipmicompare.com/, the homepage that now names the brand in its first sentence. Gemini's wording is a paraphrase of that sentence.

Claude read /insurers/, a head-to-head page, now reachable from the menu.

Newly read by any engine: /guides/international-health-insurance. No longer read: the expat comparison-tools guide, which is where Gemini's one lost answer came from.

The two "cheapest" and "best websites" questions still name nobody from this site. Those are round-up questions; engines answer them from independent articles, and the site is in none yet.

Proven

Both answers gained on the "compare quotes online" question came from engines that read the homepage after it started naming itself. The engine's own wording matches the new sentence. That is the brand fix working, on the page it was made to.

Not yet proven

That 5 of 20 holds. One answer in twenty is one twentieth of the score, and the 28 Sep readings already differed by one between runs. A second after-reading at about four weeks decides whether this is movement or jitter. It goes on this page either way.

A US marketing agency and two sites it runs · Aug 2026

The agency's own site went 56 to 89 on 2 Aug (schema, sitemap, meta description added) and 89 to 100 on 6 Aug (llms.txt). The second site went 90 to 96. The third was already 100 and was left alone. Dated before/after records on file for each.

AI answers did not move. The agency site was named in 0 of 20 on every reading from 28 Jul to 14 Aug, before and after the fixes. The third site, already 100/100, was also 0 of 20 throughout. The second site went 3 to 6 of 20 on 30 Jul, three days before any work, so the work gets no credit for it.

What it shows: passing every technical check is the floor, not the lever. Not re-measured since 14 Aug.

An independent fashion accessories shop on Squarespace · Sep 2026

Website check 82 on 1 Sep, 98 on 30 Sep. Fixed: llms.txt added, a search description added (was missing), search title lengthened from 26 to 53 characters, 97 of 98 product titles brought within the length Google shows in full, broken buy buttons and a dead contact address repaired, a spam page removed. Image descriptions are not compared because the 1 Sep check counted empty ones as present; the check was corrected, not the number.

AI answers not yet measured. The paid work was the technical fixes; the answer check is a separate step the client has not commissioned.

What it shows: one platform constraint recorded honestly (Squarespace cannot vary the heading per category page). Delivered as a completion report, 30 Sep.

A US medical practice · report only, Jul to Aug 2026

A strong starting position: named in 10 of 20 AI answers on 28 Jul and 11 of 20 from 3 Aug to 14 Aug, within one answer of each other. Website check 71/100; the main gap is a missing llms.txt. The practice took the report and made its own changes.

Nothing can be attributed. No baseline was recorded before the practice's own changes, and the website check has not been re-run since, so there is no before/after pair to score.

What it shows: without a dated baseline, a result cannot be proved, whoever does the work. This is why every engagement now starts with one.

How to read these

Two of the four moved on the website check and none has yet moved on AI answers in a way the work can claim. We publish that because it is true, and because it is the point: the technical layer is quick to fix and easy to prove, and on its own it does not get a site named.

What has been shown to move the answer count, on IPMIcompare above, is the brand named in the first sentence of a page the engines read, and answer pages that are indexed and linked. Both are content and structure work, done after the technical floor is in place.

A reading pair is evidence; a trend over several readings is proof. Every site here is due a fresh AI answer check before any further claim is made.