Ask ChatGPT to recommend an employee mental health platform for a UK company, and a certain UK wellbeing vendor - genuinely competitive in exactly that space - does not appear. Not ranked low. Absent. In my August 2026 sweep, three runs out of three returned a US-heavy roster of mental health vendors, assembled from the model's memory, without the client in it.
Ask Google the equivalent commercial query and the same thing happens: the client is absent from page 1, which belongs to national charities, the official workplace-advice body, and a handful of competitors.
Two different systems, different mechanisms, different failure modes - and the same silhouette of the same hole. That agreement is the finding. This case study is about why absence is the evidence worth building strategy on, why presence usually is not, and how this particular gap earned its place as the client's number one content target.
§ Absence beats presence as evidence
Here is the asymmetry that took me an embarrassingly long time to articulate, and now governs how I pick content targets.
When an engine mentions a client, that proves little. The run might be flattered by phrasing, by the model having recently ingested one lucky page, by the query accidentally echoing the client's own vocabulary. Presence is fragile, and worse, presence flatters - it invites you to conclude the work is done.
Absence under favorable conditions is the opposite: hard evidence. In earlier testing, some of these mental-health queries had been phrased in ways that leaned toward the client - and the engines still left it out. When a system is given every chance to name you and does not, that is not noise. It is the model telling you it has no association between your brand and that frame. And when a second, independent system - Google's page 1, built by an entirely different mechanism - shows the same hole, the gap graduates from observation to structure.
So the target-selection rule I now apply: select content targets on absence evidence, score progress on presence evidence, and never confuse the two directions.
§ Reading what fills the hole
The absence itself says "you are not here." What fills the space says why, and the two engines gave complementary diagnoses.
ChatGPT's roster was US-heavy - the big American mental health vendors, answered from model memory rather than live retrieval. That is a training-data verdict: in the corpus this model learned from, the "employee mental health platform" frame belongs to American brands, and the client's UK-specific strength never made it into the association. Google's page 1 was institutional - charities, the national advice body, plus competitors. That is an authority verdict: on this vertical, Google trusts institutions and the client has not earned a seat among them.
Same gap, two named causes, each pointing at different work: the memory gap wants extractable, citable content that future crawls and retrievals can pick up; the authority gap wants links and earned validation in the vertical. Knowing which wall you are climbing is the difference between a content plan and a content lottery.
A third signal made the target more urgent. In the client's own Search Console, long conversational queries had started appearing - full sentences like "most trusted mental health platforms by hr managers based on user reviews," with hundreds of impressions, where the client ranks positions 2 to 3. Machine-phrased searches are arriving in Google, the client already surfaces for them, and the vertical they concern is the one the answer engines will not name it in. The gap is live on every surface at once.
§ What honesty requires when the gap resists
One more measurement belongs in this story precisely because it is not a victory. A month before this sweep, a content piece engineered for this exact vertical had been published. The August runs show the gap unmoved: still 0 of 3 on the anchor query.
That is the realistic timeline, and I report it as such. Model memory updates on training cycles, not publishing schedules; retrieval pickup depends on indexing, authority and luck; a four-week-old article closing a training-data gap would have been the anomaly. The measured absence is the baseline the work is judged against - quarterly, on the same queries, with the same multi-run method - and "not yet" is a legitimate data point on the way to "yes." A consultant who cannot report "not yet" without flinching will eventually report something worse.
§ How to map your own absence
- Build a query set for each vertical you believe you compete in, and run each query multiple times per engine. Single runs produce anecdotes; frequencies produce evidence.
- Log absences separately from presences, and weight absences under favorable phrasing as your strongest signal. That list, not your keyword rankings, is your real AI-visibility gap map.
- For each absence, write down what filled the space - whose brands, what content formats, answered from memory or from retrieval. The filler is the diagnosis.
- Cross-check every AI absence against the classic SERP. A gap confirmed by two independent systems is structural and worth a quarter of effort; a gap in one engine only may be noise.
- Set the re-measurement schedule before publishing the content that targets the gap, and expect the first re-measurement to show nothing. Write that expectation down where the client can see it.
Numbers dated as of the August 2026 audit: AI results from 3-run scraper sweeps per anchor query, SERP composition from live UK pulls, search-query data from Google Search Console over a 90-day window. The site is anonymized as a matter of client confidentiality.
The reframe this method produces is the valuable part: stop asking "where do we show up?" - the comfortable question with the flattering answers. Ask "where are we missing despite deserving a seat?" That list is short, it is verifiable, and it is the only content brief the machines themselves have already approved.