Case study · Reputation
They won page one.
The AI answer never noticed.
Reputation work is measured on the first page of results. The AI answer above those results is built from a different set of sources — and pushing a page down does not remove it from that set.
An enterprise deep-tech company selling AI simulation software into pharma, defense, and financial services, carrying unfavourable third-party coverage on both its company and executive brand queries. Measured on both surfaces across a four-month programme.
7 days
Publish to news tab
From the first article going live to first impressions in Google's News tab.
1 → 2
Owned results, page one
Owned properties on page one of the brand query, displacing third-party results.
5th
Own site as a source
Where their own domain ranks among sources AI cites in answers about them.
4 of 5
Answers carrying it
Sampled AI responses still repeating the unfavourable narrative.
What worked
Conventional reputation work did its job.
A news sitemap where none existed. An owned editorial stream published on a regular cadence. Press releases structured so machines could extract facts from them cleanly — compressed headlines with entities preserved, definition pairs, bulleted incident lists, canonical links on first mention only. It moved the results.
Google Search Console · News search type · owned editorial path
By conventional measurement, this is a successful reputation programme.
What page one didn't reach
The answer draws from somewhere else.
During the same period an AI answer appeared directly in the branded search result, replacing the social carousel that had occupied that position. The question "who is this company" stopped being answered by a list of links.
That answer is assembled from a source set of its own. We measured it. The unfavourable third-party coverage that page-one work had pushed down was still being cited as a source in AI answers about the company — sitting alongside a major business publication, feeding a paragraph most buyers now read before they read anything else.
And the company's own domain was not the primary source about itself. It ranked fifth, behind a reference encyclopedia, a business publication, a university news page, and a personal biography site.
Two surfaces, one brand, measured in the same window
Why this happens
Ranking is a position. Citation is a decision.
Displacement works because search results are ordered — publish enough credible owned material and unfavourable coverage moves down the list. An AI answer is not a list. It selects sources and summarises them, and a source that is no longer on page one can still be in that selection. Which is why the narrative persisted: sampled AI responses kept repeating the unfavourable framing after the page-one work had visibly succeeded — and nothing in a rankings report would have shown it, because the rankings had improved.
A ranking is a position
Ordered. Add enough credible material above an item and it moves down, out of sight.
A citation is a decision
Unordered. The model picks what it will quote — and rank on the page has no bearing on that choice.
You can win the page and lose the paragraph.
Reading the numbers
What these figures do and don't say.
Stated plainly
- Displacement was partial, not total. One unfavourable third-party item remained on page one of the executive query at the final measurement. An earlier internal summary claimed full clearance; re-checking the captures showed that was wrong, and it was corrected. We publish the corrected version.
- The AI-source finding is an observation, not an attribution. That a specific page is cited is directly measurable. Why an AI model selects one source over another is not, and nothing here claims to know.
- Coverage is Google and ChatGPT. Measurement across other assistants was attempted and returned near-zero — but a second vendor reported meaningfully different figures for the same period. Treat any single-vendor AI-visibility number, including ours, as one vendor's view rather than ground truth.
- Search and news figures are first-party, from the client's own Search Console, filtered to the relevant paths and dates.
Start here
Find out what the answer says about you.
Your page one may look fine. That is not the same as the paragraph above it being fine.
We measure both — what ranks for your brand, and which sources the AI answer is actually built from. You'll see the same comparison shown here, about your own name.