If ChatGPT, Gemini or Perplexity gets facts about your business wrong, you cannot edit the answer directly, but you can usually fix it at the source. The method is: ask the assistants a fixed set of questions, record what they say, trace each error to the page or listing it most likely came from, correct that source, and re-test every few weeks. Most wrong answers trace back to outdated listings, thin or contradictory website copy, or a missing page, and all of those are fixable.
This matters because buyers increasingly ask assistants things like “who does X in my area” or “is Company Y any good” before they ever visit a website. A wrong opening hour is an annoyance. A wrong price, a service you no longer offer or a confusion with a similarly named company costs you enquiries you will never know you lost.
Why do AI assistants get businesses wrong?
Assistants answer from a mix of training data and, increasingly, live retrieval from the web. Errors usually come from one of five places:
- Stale information. Old pricing, a previous address, a discontinued service, copied across directories years ago and still repeated.
- Contradictions between your own sources. Your website says one thing, your Google Business Profile another, a directory a third. The model has no way to know which is current.
- Name collisions. Another company, person or product shares your name, and the assistant blends them.
- Gaps. If there is no clear page stating a fact (your pricing, your service area, who you work with), the assistant may guess from similar businesses.
- Weak or untrusted sources. Third-party pages that rank or get cited for your name, such as old reviews, scraped directories and forum threads, can outweigh your own site.
Step 1: Run a 20-question audit
Open ChatGPT, Gemini and Perplexity (and Copilot if your audience uses Microsoft products). Use a clean session, not logged-in history that might bias the answer. Ask each assistant the same set of questions and record the answers in a spreadsheet with the date. A good starter set:
- “What is [Company Name]?” and “What does [Company Name] do?”
- “Where is [Company Name] located and what areas does it serve?”
- “How much does [your main service] cost with [Company Name]?”
- “Is [Company Name] a good choice for [your typical customer]?”
- “Who owns or founded [Company Name]?”
- “What are the best [your category] providers in [your market]?”
- “Alternatives to [Company Name]” and “[Company Name] vs [main competitor]”
- “What do customers say about [Company Name]?”
- Three or four questions your customers actually ask before buying, phrased the way they would phrase them.
Repeat each question once or twice. Assistants vary between runs, so a single wrong answer is a signal, while a consistently wrong one is a pattern worth fixing first. If you want to keep tracking this over time, our guide to monitoring brand mentions across the web describes a lightweight setup.
Step 2: Classify each error
Score every wrong or missing answer into one of four buckets:
| Type | Example | Likely fix |
|---|---|---|
| Factual error | Wrong price, location, service list | Correct the source page and listings |
| Omission | Your business never appears for key questions | Create or strengthen the page that answers the question |
| Confusion | Mixed up with a namesake | Disambiguation: clearer entity signals |
| Negative or outdated reputation | Repeats an old complaint | Fresh reviews, an updated public response |
Step 3: Trace the error to its source
Perplexity and some modes of ChatGPT and Gemini show citations. Click them. Where an answer appears without sources, search for the wrong fact in quotation marks on Google and Bing. In our experience the origin is frequently one of the following, in rough order of frequency:
- Your own website, where a forgotten page or an old blog post still states the old fact.
- Your Google Business Profile or other directory listings.
- A third-party review site, aggregator or “best of” article.
- A forum thread or social post. See our note on why answer engines lean on Reddit for why those matter.
Step 4: Fix the sources, in this order
1. Your own website first
Make sure one page states each important fact clearly, in plain sentences: what you do, who it is for, where you operate, what it costs or how pricing works, how to contact you. Remove or redirect obsolete pages. Add a short “About” statement that begins with a plain definition (“Company X is a …”). Our piece on formatting a pricing page so assistants quote it correctly applies directly if the error is about money.
2. Profiles and listings you control
Update Google Business Profile, Bing Places, LinkedIn, industry directories, and any marketplace listings. Make name, address, phone, description and category identical everywhere. Inconsistency is how assistants end up unsure.
3. Structured data and entity signals
Add Organization schema on your homepage with your legal name, logo, URL and links to your official profiles through the sameAs property. If you share a name with others, add a brief disambiguating phrase in your descriptions: city, sector, founding year. We explain the background in our guide to entity SEO.
4. Third-party sources you do not control
Contact the owner of the page and request a correction; most directory and article owners will update a factual error if you send the correct information and a link to proof. For outdated forum posts, a clear, honest reply from an official account with the current facts can be enough. Do not argue; state the current facts once, politely.
5. Use feedback tools sparingly
Most assistants offer thumbs-down or report buttons. They are worth using for clear factual errors, but treat them as supplementary: there is no guarantee that a single report changes future answers, and fixing the underlying sources is what lasts.
How long does it take for corrections to show up?
Assistants that retrieve live web results can reflect an updated page within days or weeks once it is crawled. Parts of the answer that come from model training data change only when the model is updated, which you cannot control. Expect a mixed picture: some answers improve quickly, others lag for months. That is a reason to start now and re-run the audit every four to six weeks, recording the dates.
A worked example (hypothetical)
Imagine a small accountancy firm that stopped offering payroll services two years ago. An assistant still lists payroll in a summary of the firm. The audit shows it on all three assistants. Tracing it, the firm finds an old directory listing and a forgotten service page. The fix: delete the page with a redirect to the current services overview, update the directory entry, and state in the About page and Business Profile exactly which services are offered. Three weeks later a re-test shows one assistant has corrected itself and two have not. The firm adds a short FAQ (“Do you offer payroll? No, we focus on …”) and tests again after a month. The point is the process, not the result: observe, trace, correct, re-test.
Prevention: keep your facts in one place
The cheapest long-term fix is a single “facts page” or a section in your About page that holds the canonical answers: services, areas, pricing approach, founding year, team, contact details. Whenever something changes, update that page first, then your listings. Put a review reminder in the calendar each quarter. When your facts are consistent and findable, assistants have less room to improvise, which is also what makes a business quotable in the first place. If you want a broader checklist, see our eight-check GEO audit.