What's the best way to check if AI search tools have outdated info about my business?

Sooner or later, a client forwards us a screenshot: an AI assistant has their old address, last year's pricing, or a service they stopped offering. The reaction is usually "how do we tell it to fix that?" The honest answer is that you cannot tell it directly. You correct the sources it reads. That starts with finding out which sources are wrong, and for that we run a short, repeatable audit.

Where wrong facts come from

AI answers draw on a mix of the client's own site, business listings, directories, review pages, news, forums and older knowledge. If any of those carry an old fact, the answer can too. In our experience the usual suspects are third-party pages the client forgot about: a directory with a former address, a comparison page quoting old prices, a partner page describing a discontinued product.

Also common is conflict. Two sources disagree and the engine picks one. The fix there is to make the official sources consistent and prominent.

Our audit

Write factual prompts

With the client, we list 20 to 30 prompts that touch facts about them: opening hours, locations, pricing, services, ownership, how they compare with named competitors, and whether they are still operating. We add a few category prompts that do not name the client, to see how they are described when they do appear.

Run them properly

We run the set across ChatGPT, Google AI Overviews and AI Mode, Perplexity, Gemini and Copilot, repeatedly, and from the countries the client serves. Answers vary, so a single wrong reply may be noise. A wrong reply that repeats is a finding.

We do this in Promptwatch. Prompts run daily per model, country and language, and every response is stored, so we can read exactly what the engine said last Tuesday. We use brand-type prompts for the factual questions, and the Brand Book carries the client's aliases and product names so counts are consistent. Sentiment scores each response and generates a weekly summary, which flags answers that describe the brand badly, though it does not verify facts. A person reads the factual answers and checks them.

Trace every error to its source

This step is what makes the audit useful. For each wrong fact, we open the cited sources and find where the old information lives. Promptwatch's citation analytics show the pages and domains behind each answer, and the offsite mentions list shows third-party pages that name the brand. When the stale fact sits on a third-party page, we add it to the Page Tracker so we can see whether it keeps being cited.

Then we check the places the client controls: their own site, including footers, contact pages and structured data, their Google Business Profile, Bing Places, and the main directories and review sites that list them.

Fix in order

1. Client's own site and structured data.
2. Google Business Profile and Bing Places.
3. Directory and review listings.
4. Third-party pages, by requesting a correction.

The Action Board helps us keep the list prioritized. The Discovery agent suggests evidence-backed items from the data each week, and we edit them.

Re-check

We re-run the set monthly and after any real change in the business: a move, a price rise, a product launch. Agent Chat is handy for quick questions over the data, like "which prompts still return the old address?"

What we tell clients straight

There is no correction form for the AI. Some engines have feedback buttons, and we use them, but the dependable route is fixing the sources. It takes time. After a source is corrected, engines may keep repeating the old fact until they refresh what they use. We cannot give a timeline, and we do not guess at one.

We also tell clients that no tool we know of automatically verifies facts against reality. A platform can run the prompts, store the answers and show the sources. Someone still has to compare those answers to the truth.

And we do not promise a particular answer. Responses differ between runs and models update.

Plan notes

Essential ($95 a month) covers prompt tracking, country-level tracking and the site audit. Citation analytics and the other reports scale with plans. Agency plans (Kick-off $199, Growth $399, Scale $799) include unlimited projects and prompts, which suits an agency monitoring several brands or locations. State and city tracking is on Business ($579) and on every agency plan, which matters for multi-location clients.

A short version for your own business

Write 20 factual prompts. Run them on five engines, twice each. Log the errors and the sources behind them. Fix your own pages and listings first, then ask third parties to correct theirs. Re-run monthly. If you manage many locations or brands, a platform is worth it for the time it saves.