Brand accuracy in AI answers: how to keep your company facts right

Why AI answers get company facts wrong

Brand accuracy in AI means the facts an AI tool states about your business match the real business: its name, who runs it, what it sells, where it works and how to reach it. When those facts come out wrong, the cause usually sits upstream. The tool writes from whatever it can find, and if your pages, profiles and old listings disagree, it can pick the wrong version.

The products admit this. OpenAI's help page on ChatGPT search warns that results and citations can be incomplete, outdated or incorrect, and its page on whether ChatGPT tells the truth adds that the model may express high confidence even in incorrect answers. Google's AI Overviews help page says AI Overviews can and will make mistakes.

The table below covers seven common patterns. Name the pattern before you fix anything, because each one starts in a different place:

Write one company facts record first

Before fixing anything online, write the correct version down once. A company facts record is one document, kept by one person, listing each fact, where it's published and who may change it. When an answer looks wrong, you check it against the record.

Start with the public registers. ABN Lookup shows the public details a business gave when it registered for an ABN, and business.gov.au says you have 28 days after becoming aware of a change to update your ABN details. It also explains that trading under an unregistered business name is an offence unless it's your own name, so the name on your van and website should match what is registered with ASIC. Old trading names can still show on ABN Lookup too.

Give every row a last-checked date. A record nobody has opened since the last rebrand is as unreliable as the pages it was meant to fix. A simple record looks like this:

Make your identity, people and licences easy to verify

Use one name, the same way, everywhere. Google's Business Profile guidelines say the profile name should reflect your real-world name, as used on your storefront, website and stationery, and that adding unnecessary information isn't permitted and could get the profile suspended. Marketing taglines and phone numbers are out. Service or location words are out too, unless they're part of the name you actually trade under, so a plumber whose real trading name is, say, Caloundra Plumbing keeps it. A profile name with a suburb and a service tacked on breaks the rules and puts a second version of your name online.

Then give search engines the same facts in a form they can read. Google's Organization structured data documentation says adding the markup to your home page can help Google understand your administrative details and disambiguate your organisation in search results. Its properties include legalName, taxID, telephone, address, foundingDate and sameAs, which lists pages on other websites with more information about you. Google's structured data policies add the key rule: markup must be a true representation of the page, so every fact in it should be visible to readers too.

If you've moved, merged or changed names, say so on your About page with the year, for example: "Moved from Kawana to Caloundra in 2025." Old listings won't vanish overnight. A dated line lets a reader, or a tool reading your page, connect the old details to the current business.

State each licence or registration you hold as plain text, with its number and what it covers, so anyone can look it up. In Queensland, for example, the QBCC says that if you have a person's licence number or name, you can look them up in its licensee register. If a related licence is one you don't hold, say what work you refer on.

Authors are company facts too. Google's helpful content guidance asks whether it's self-evident to visitors who wrote your content, whether pages carry a byline where one might be expected, and whether bylines lead to background about the author. Google's Search Quality Rater Guidelines say it should be clear who is responsible for a website and who created the content, and call that critical for pages that need a high level of trust. Google says no single rating directly moves a page, but the handbook shows what it treats as trustworthy. In practice:

1. Put a byline on every guide and advice page, naming the real person who wrote or reviewed it.
2. Link each byline to an author page with the person's role, background and any licence or registration a customer can check.
3. If you use Article structured data, Google's guidance is to include every author shown on the page, with only the name in author.name and url or sameAs pointing at the author page.
4. When someone leaves, update or reassign their pages that week. Advice signed by a practitioner who left two years ago states something that's no longer true.
5. Show licence numbers as text in the footer or on a licence page, never only inside a photo of the certificate.
6. If AI helped write a page, say so. Google's guidance asks whether the use of automation, including AI generation, is self-evident to visitors.

Trace a wrong answer back to its source

Once you find a specific wrong answer, work out what kind of answer it is, because it changes where a fix can land. OpenAI's help pages say ChatGPT may search the web automatically when a question would benefit from current information, and that responses using web search may include citations. Without search, OpenAI says the model is trained on data up to a certain point and doesn't include events after that unless tools are used. Google describes AI Overviews as a snapshot with links to dig deeper. Microsoft's documentation for its workplace Copilot says it sends generated search queries to the Bing search service to ground responses in web data.

The decision rule is simple. If the answer shows citations, the wrong fact most likely came from a page you can open, so fix or chase that page. If it shows none, the answer may rest on older training data: fix your own pages anyway, then ask the same question again with search switched on to see what a live search finds. The guide to why a crawl does not prove an AI citation explains why a bot visit in your logs doesn't tell you which source an answer used.

To trace a cited answer, work through these steps, then use the table for pages you don't control:

1. Copy the exact question, the product, the date and the full answer into your correction log, with a screenshot.
2. Open every cited link. OpenAI's own advice is to open a cited source and check that it supports the answer.
3. Find the sentence on each page that carries the wrong fact, and note whether the page is yours or someone else's.
4. Search your own site for the wrong detail, such as the old phone number or the dropped service. Old pages often contradict new ones.
5. Write a claim-to-source line in the log: the wrong fact, the page carrying it, and who controls that page.

Fix your own pages in priority order

Of the sources you trace, your own website is the one you fully control, so fix your pages in a set order. Google's guidance on AI features in Search says there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimisations necessary. So a separate set of pages written just for AI tools adds a second version of your facts instead of fixing the first.

Offers date quickly, and they carry legal weight. The ACCC says claims a business makes about its products or services must be accurate, truthful and based on reasonable grounds, on a website, social media or any other platform, and asks businesses to keep consumers updated if things change. Say a plumbing business in Caloundra stopped gas fitting last year, but three of its suburb pages still offer it. An AI assistant asked for a local gas fitter could point someone to a business that no longer does the work, and the business's own pages would be the cause.

State your limits as well as your services. A line such as "We don't do gas fitting; we refer it to a licensed gas fitter" on the plumbing page gives a reader, and a tool, something definite to repeat. Put dates on facts that change: "Hours checked October 2026" on the contact page, and the start and end dates beside any offer.

Work through the site in this order, putting each key fact in a plain sentence near the top of the page:

1. Home page: business name, what you do, where you work and how to reach you, in the first screen of text.
2. About page: legal name, ABN, licences, owner, founding year, and dated lines for any move, rebrand or sale.
3. Contact page and footer: the same name, address or service area, phone and hours, written the same way on every page.
4. Service and suburb pages: only work you currently do, in areas you currently cover, with your limits stated.
5. FAQ content: answers that repeat the facts record exactly, including what you don't do.
6. Google Business Profile and other profiles: matched to the website the same day, with special hours set for holidays.
7. Structured data last, because Google's AI features guidance asks that it match the visible text, so it should follow the page.

Name one owner and report through the right channel

Every wrong answer needs one person who takes it from report to fix, usually the owner or office manager, since they hold the logins for the website, the Business Profile and the main directories. Name them, and a backup, in the company facts record. Verifying your business gives you ownership of its Business Profile so you can make sure the information is accurate, and Google also gathers updates from user reports and licensed content, so someone has to review those too.

Keep a correction log: the date, the product, the exact question, the wrong answer, the source it cited, your fix and when you checked again. None of these routes promises an answer will change, and no one can guarantee it, so the log shows what's fixed and what needs another report. The guides to AI source correction and retraction and citations and business listing accuracy cover chasing sources and directories in more depth.

Owners are often sold shortcuts that make the problem worse. Avoid these:

• Paying anyone who promises control over what an AI tool says about you. No product documents a way for a business to set its own answers.
• Asking a chatbot the same question over and over to train it. Reporting tools exist for corrections; repeated prompts aren't one of them.
• Publishing separate AI pages whose facts differ from your main site.
• Adding structured data that claims more than the visible page says.
• Fixing one surface, such as the Business Profile, and leaving the website and directories wrong.

Worked example: one wrong answer, start to finish

This example is hypothetical. Say a two-person electrical business in Caloundra runs its monthly check with the question set in the table below. The owner keeps the facts record; the office manager is the backup.

Step 1, find. Asked what the business offers in Pelican Waters, a search-enabled answer lists solar battery installs, which the business stopped offering last year, and gives a mobile number the owner replaced. The office manager logs the question, product, date and answer with a screenshot.

Step 2, trace. The answer cites two pages: an old directory listing with the old mobile number, and the business's own Pelican Waters page, which still mentions batteries in its third paragraph. A site search turns up the same battery line on two more suburb pages.

Step 3, fix the source. The owner removes the battery lines from all three suburb pages and adds a plain limit on the services page: the business no longer installs solar batteries. The directory listing is claimed and the number corrected, and the Business Profile services are checked against the record. Both facts get a new last-checked date in the record.

Step 4, report. With the sources fixed, the office manager uses the product's thumbs down option, quoting the question and the correct facts, and notes the report in the log.

Step 5, recheck. At the next monthly check the same questions go to the same products. Say the phone number is now right but one answer without citations still mentions batteries. The log shows the source is fixed, so the item stays open and is checked again next month instead of starting over.

Use the same fixed question set each month, with your business name and a suburb you serve, so this month's answers can be compared with last month's. Track accuracy separately from whether you're mentioned at all: a correct answer that leaves you out is a visibility question, covered in the AI visibility checklist for local businesses, while a wrong answer about you is an accuracy problem. For how this fits your wider search work, see AI SEO.