In May 2026, a UK employee wellbeing platform had a comparison article - "the best platforms in our category, and how we stack up" - that was live, well-written, and completely invisible. Google Search Console filed it under "Crawled - currently not indexed." It was not in the sitemap. No engine cited it. For practical purposes it did not exist.
In August 2026, the same article was ChatGPT's number one cited source on the core category query - "best employee wellbeing platform UK" - in 3 out of 3 test runs. ChatGPT positioned the client "Best overall" in all three, sourced from the client's own page. In one spot check, it called the piece one of the "independent UK comparisons" it had consulted.
Nobody wrote new content to get there. The best-performing AI asset this company owns was already sitting on the server. This case study is about why it was invisible, what changed, and the uncomfortable footnote about what "independent" means to an answer engine.
not indexed
0 citations, not in sitemap
ChatGPT's #1 source · Google #6
§ Why a perfectly good article did not exist
The article had two problems, neither of them about quality.
First, discovery. It was not listed in the sitemap, so the one file that tells crawlers "here is everything we have" did not mention the site's most strategically important page. Second, indexing. Google had crawled it and declined to index it - the "Crawled - currently not indexed" state that usually means the page has no internal links pointing at it, no external signals, and nothing telling the crawler it matters.
That combination is worth naming as a category: the invisible asset. Content that has already absorbed the expensive part - research, writing, positioning - and is leaking all of its value through the free part: discovery plumbing. In my audits it shows up more often than you would guess, because content teams measure what they publish, and nobody measures what the sitemap forgot.
§ What actually changed
The fix list was unglamorous: put the article in the sitemap, give it internal links, rebuild it as a proper citation piece with question-form headings and answer-first paragraphs, and let the crawlers re-encounter it.
By the August re-audit, the results were measurable on both surfaces. On Google, the article ranked #6 on the category query - mid page 1, sitting among the third-party listicles. On ChatGPT, it had become the citation engine for the whole category cluster: the primary source on the "best platform" query 3 of 3 runs, cited on the enterprise variant, cited on the how-to-choose query alongside a second owned article, cited in a head-to-head comparison against a named competitor.
The mechanism is the one I keep finding across engagements: on "best X" queries, answer engines cite comparison-format content - somebody's listicle. If the category's publishers have not written a good one, the field is open for a vendor to become the listicle. This client did, and displaced the third-party roundups in its own category answer.
§ Two footnotes that keep the story honest
The engines have noticed who wrote it. In the head-to-head comparison runs, ChatGPT twice added a caveat: the comparison content is authored by the vendor it recommends, and "I wouldn't treat its claims as independent evidence." Cited 3 of 3, recommended, and labeled. Vendor-authored comparison content still works - these runs prove it - but the engines are learning to flag it, which means third-party validation is not optional forever. Earned citations remain the moat; owned comparisons are the beachhead.
A duplicate leaked into the citations. In one run, ChatGPT cited the article via its locale-duplicate URL - a second copy of the same page under a language prefix, with no canonical tag telling machines which copy is real. Citation equity splitting across duplicate URLs is exactly the kind of erosion that never shows up until you read the citations line by line. The canonical fix went on the queue the same day.
§ How to find your own invisible assets
- Diff your sitemap against your actual published content. Anything live but unlisted is leaking. This client's most important article was in that gap.
- In Search Console, read the "Crawled - currently not indexed" list as a triage queue, not a curiosity. Sort by what the page would be worth if it worked.
- Check whether your category's "best X" query has a strong third-party listicle. If it does not, the citation slot is open, and your own honest comparison can take it.
- Once a piece starts earning citations, read the cited URLs exactly. Duplicates, locale copies and parameter variants in the citation list are signal fragmentation you can fix with canonicals.
- Track the same query on a schedule with repeated runs, not single spot checks. One run is an anecdote; 3 of 3 is a claim you can put in writing.
Numbers dated as of the May and August 2026 audits: citation results from 3-run scraper sweeps per query in August 2026, indexing states from Google Search Console, rankings from live UK SERP pulls. The site is anonymized as a matter of client confidentiality.
The summary that should bother you: the most valuable AI-visibility asset this company owns spent months invisible for the price of a sitemap entry and some internal links. Before you commission the next content calendar, find out what you have already paid for that no machine has ever been shown.