To check what AI assistants say about your business, ask each assistant the questions a real customer would ask, in a clean session, and record the answers and the sources they cite. Do it with a fixed list of questions so you can repeat the check every month and see what changes. The whole process takes about 30 minutes the first time.

Below is the method we use, a template for recording results, and what to do when an answer is wrong.

Step 1: Write your question list

Start with 10 to 20 questions in three groups. Write them the way customers actually talk, not the way you describe yourself.

Brand questions: does the assistant know you?

  • "What is [Company name]?"
  • "Is [Company name] a good [service provider]?"
  • "Where is [Company name] located and what do they do?"

Category questions: do you appear when someone does not know your name yet?

  • "Best [service] in [city or region]"
  • "Who can help with [specific problem]?"
  • "[Service] for [type of customer]"

Comparison questions: how are you described next to alternatives?

  • "[Company name] vs [competitor]"
  • "Alternatives to [competitor]"

For a tax firm, a category question might be "accountant for cross-border tax in New Jersey". For a B2B supplier, it might be "peptide API supplier with GMP documentation". Use your sales conversations: the questions prospects email you are the questions they now ask an assistant first.

Example lists by type of business

The best questions are specific to how your customers buy. Three starting points:

A professional services firm (tax, legal, accounting, consulting):

  • "Accountant who handles [specific situation] in [city]"
  • "What does [Firm name] specialize in?"
  • "Do I need a CPA or an enrolled agent for [problem]?"

A local service business (contractors, clinics, home services):

  • "Roof repair near [town]"
  • "Is [Company name] licensed and insured?"
  • "How much does [job] cost in [region]?"

A B2B manufacturer or supplier:

  • "Suppliers of [product] with [certification or specification]"
  • "[Company name] product range and minimum order"
  • "[Company name] vs [competitor] for [application]"

Notice that several of these are questions a website can answer directly: services, locations, credentials, process. If your site does not answer them, an assistant has to find the answer somewhere else, or guess.

Step 2: Use clean sessions

Assistants personalize answers based on your history, location and settings. You want to see what a stranger sees.

  • ChatGPT: use a temporary chat, or a browser where you are signed out, and make sure web search is used for current questions.
  • Perplexity: use a signed-out or private window. Perplexity shows its sources next to every answer, which makes it the easiest place to start.
  • Google: search in a private window. AI Overviews appear only for some queries; Google says they show when its systems think they add something to regular results, so note when one does not appear at all.
  • Microsoft Copilot: use a signed-out session. Its web answers are grounded in Bing results.
  • Gemini and Claude are worth adding if your customers use them.

Ask each question once per assistant, then ask one or two again. Answers vary between runs, so treat a single answer as a sample rather than a verdict.

Step 3: Record what you see

A spreadsheet is enough. One row per question and assistant:

ColumnWhat to write
DateWhen you checked
AssistantChatGPT, Perplexity, Google AI Overview, Copilot…
QuestionExactly as asked
Mentioned?Yes / No / Only when named
AccuracyCorrect / Partly wrong / Wrong
What it saidShort summary or quote
Sources citedThe URLs the answer links to
Competitors namedWho appears instead of or alongside you
ActionWhat to fix, and where

A filled-in row might read: October 13 · Perplexity · "accountant for cross-border tax in New Jersey" · Not mentioned · — · Three firms listed, sources were two directory pages and one firm's service page · Action: add a dedicated cross-border tax page that names the state and the situations it covers. One row like that tells you more than a page of rankings.

The Sources cited column is the most useful one. It shows which pages the assistant relied on. Often they are not your website. They are a directory listing, an old press release, a review site or a competitor's comparison page.

Step 4: Diagnose the gaps

When you review the sheet, most problems fall into four patterns.

You are not mentioned at all. For brand questions, this usually means the assistant cannot find or read your website. Check crawler access and whether your content is in the HTML (see What Makes a Website Visible to AI Search?). For category questions, it usually means no page on your site clearly answers that question.

The facts are wrong. Old addresses, discontinued services and wrong descriptions almost always come from a source you can find. Look at the cited URLs. If it is your own site, update the page. If it is a listing or directory, correct it there.

The description is vague. "A company that offers various services" means the assistant found nothing specific. Add plain, concrete statements to your homepage and About page: what you do, for whom, where.

Competitors appear instead of you. Read the pages the assistant cites for them. They usually answer the question more directly, list specific services or locations, or appear on the comparison and directory pages the assistant trusts.

Step 5: Fix the source, then re-check

Assistants summarize what they find. You cannot edit their answers, but you can change what they find:

  1. Correct your own website first: company description, services, locations, contact details and any outdated pages.
  2. Update the profiles you control: Google Business Profile, Bing Places, LinkedIn and industry directories.
  3. Add or improve pages that answer the category questions where you were missing.
  4. Make sure new pages are crawlable and listed in your sitemap.

Then wait. Crawlers need to revisit the pages, and retrieval systems need to pick up the change. Re-run the same questions after a few weeks and compare rows.

How often to check

Monthly is a good rhythm for most businesses. Check sooner after anything that changes your public facts: a move, a rebrand, a new service line, a website rebuild. Keep the question list stable so the results stay comparable. Add questions over time, but don't rewrite the old ones.

Who should run it

Give the check one owner. It does not need to be a specialist, but it does need to be someone who knows the business well enough to spot a wrong answer: an outdated service, a price you no longer charge, a competitor described as doing what you do. In a small company that is often the founder or whoever answers new inquiries. Keep the sheet in a shared place, and when you change something on the website or a listing, note it in the Action column with the date. After a few months, that column becomes a record of what you fixed and what changed as a result.

What this check can and cannot tell you

A manual check shows what assistants say to a stranger today, for the questions you chose. It cannot tell you how often those questions are asked, and it cannot guarantee tomorrow's answer. Treat it like a regular inspection, not a ranking report. Its value is in spotting wrong or missing information early, and fixing it before customers act on it.

On our annual plans, this monitoring is built in: we track a fixed set of buyer questions across the main assistants, report what changed and fix what we find on the website.

Frequently asked questions

Can I ask ChatGPT or Perplexity to correct information about my business?

You can send feedback on individual answers, but the lasting fix is to correct the sources those assistants rely on: your website and the listings and profiles they cite.

Why does the same question give different answers?

AI answers are generated fresh each time and may retrieve different pages. Personalization, location and model updates also play a part. That is why a fixed question list, clean sessions and repeated checks matter.

Are there tools that automate this?

Yes. Several AI visibility tools run question lists across assistants on a schedule. They save time at scale, but the method is the same: fixed questions, recorded answers, cited sources and follow-up fixes.

How many questions should I track?

Start with 10 to 20 across brand, category and comparison questions. That is enough to spot patterns without making the check a chore.

Sources