Making a National Brand the AI Answer
How I corrected years-out-of-date AI answers about a national brand and got its current facts cited across the major AI engines. This case study is anonymised.
The AI answers were describing a brand that no longer existed.
Ask ChatGPT or Gemini about the brand and you got a picture that was roughly four years out of date. The positioning was old. The pricing was old. Products that had been retired were still surfacing as the headline answer. Current products, the ones the brand actually sells now, did not appear.
The brand had no visibility into any of it. Nobody was checking what the AI engines said, so nobody knew how far the answers had drifted from reality. This case study is anonymised: I refer to the subject only as a national brand.
Three reasons the engines were stuck in the past
When I audited the brand's entity footprint the way an AI engine reads it, I found three problems feeding the stale answers:
Wikidata, Wikipedia, the Knowledge Panel, and several third-party profiles all still described old positioning and retired products. These are exactly the sources the engines lean on, so the answers inherited the drift.
The site had no clean structured data for the current products. Without Organization, Product, and Offer schema stating the facts plainly, the engines had nothing current to prefer over the old third-party copy.
There was no llms.txt and no deliberate stance on AI crawler access. The engines had no signposted, canonical source of current facts to pull from, so they kept reaching for whatever they had cached.
Rebuild the entity, then make the correction measurable
This is AEO (Answer Engine Optimisation) and GEO (Generative Engine Optimisation) work: get the engines to describe the brand accurately, then prove it moved. I started with a full entity audit across Wikidata, Wikipedia, the Knowledge Panel, and the main third-party profiles, mapping every place the old facts were coming from.
Then I rebuilt the JSON-LD across Organization, Product, and Offer schema so the current facts were stated cleanly and consistently. I implemented llms.txt and set deliberate AI crawler access so the engines had a signposted canonical source. Where old snippets lived on third-party sites, I raised refresh requests to get the stale copy corrected at source.
Finally, I built a GA4 reporting layer for AI referral traffic. For the first time the brand could see visits arriving from the AI engines and track the correction week by week, rather than guessing whether anything had changed.
Current facts cited, and the change is now visible
The stale answers were corrected across the major engines. The current products started appearing in ChatGPT, Gemini, Perplexity, and Google AI Overviews where the retired ones used to. AI referral traffic showed up in reporting for the first time, so the brand can now watch its AI visibility instead of assuming it.
If you do not manage the AI answer, the AI answer manages you
AI visibility is not a vanity check. The engines are answering questions about your brand today, and they will happily do it with facts that are years out of date. Fix the entity, make the current facts machine-readable, and put a measurement layer on it. Otherwise the first thing a buyer hears about you is wrong, and you never even see it.