A rumour is circulating that you closed your Klang office. You cannot phone the rumour. You can only be so visibly, consistently open — same address on every door, every letterhead, every listing — that repeating the rumour becomes obviously wrong to anyone who checks.
Correction is not a request. It is out-evidencing the error.
Why it happens
Three causes, and the fix differs for each:
- Gaps. The model has no good information about you and generates the statistically plausible thing instead. Confidently. This is the most common cause for small businesses and the easiest to fix.
- Contradictions. Different sources say different things — an old address on a directory, a former service on a partner's site, a discontinued price in a cached PDF. The model picks one, sometimes the wrong one.
- Staleness. The information was true. You changed. The internet has not caught up, and neither has the training data.
OpenAI, Google and Anthropic do not operate a brand-correction portal. There is no ticket to raise, no dashboard, no escalation path. Anyone offering to "get your AI listing corrected" through a contact at a platform is describing something that does not exist.
The correction method
Which engine, which prompt, what it said. Screenshot it. You need a baseline or you will never know whether it improved.
Perplexity and Google AI Overviews show their citations directly — start there. For ChatGPT, search the web for the wrong claim itself; there is usually a page saying it. Very often it is your own site: an old page, a stale PDF, a legacy subdomain nobody remembers.
Update, correct, or remove the page. Then redirect it if you removed it. Roughly half of brand hallucinations trace back to something the company itself published years ago and forgot.
Directories, partner sites, old press. Email and ask. Most will fix it; it costs them nothing. This is unglamorous work and it is the work.
A page on your own site that states the truth in plain, quotable language. "DNE Total operates from Klang, Selangor. We do not provide permit advisory services." Direct statements, easy to retrieve, hard to misread.
Update every profile. Refresh your Wikidata entry if you have one. Update your Google Business Profile. Get the correct version stated somewhere other than your own site. This is Chapter 5 applied as damage control.
Same prompt, clean session. Changes propagate over weeks or months, not days, because indexes refresh on their own schedule and some of the wrong information may sit in training data that will not update until the next model.
The high-cost errors, in order
Not all inaccuracies matter equally. Fix in this order:
- Contact details and location. A wrong phone number costs you calls today.
- Services you do not offer. Attracts unqualified enquiries and, in regulated sectors, can create real liability. If you do not advise on permits, a model saying you do is a problem.
- Pricing. Stale prices poison negotiations before they start.
- Confusion with another company. Similar names get merged. This is an entity problem — Chapter 11 is the cure.
- Tone and positioning. Being described as a budget option when you are premium. Annoying, slow to fix, lowest priority.
Ask three AI tools "what does [your company] do, where are they, and what do they charge?" this week. Write down every wrong answer. That list is more urgent than any content plan.