Showing up in AI answers what a small business can actually measure and do
Nobody can guarantee a mention, and answers vary from one run to the next. There’s still plenty to measure and improve, as long as you read the results with care.
The Oryon team4 min read
Customers now ask ChatGPT, Gemini or Perplexity to suggest a supplier, and a small business is right to wonder whether its name comes up. The honest answer holds more uncertainty than many offers on the subject let on, so let’s start there.
What nobody can guarantee
- Answers vary.Ask the same question twice and you can get two different lists. Wording, model, language, location and the person’s history can all play a part.
- You can’t see the questions.For Google, Search Console shows the searches that lead to your site. Assistants generally offer nothing comparable: you don’t know what people ask them, or how often.
- Nobody controls the mention.How the models work inside isn’t public. Anyone promising you a spot in the answers is promising something they don’t control.
So treat this as a signal to watch over time, with the caution you’d give a survey of a small sample.
What you can measure with a spreadsheet
A homemade check, repeated every month, will already teach you a lot.
- Write about fifteen questions.The ones a customer would ask without knowing your name, plus a few that use it. For an accounting firm, for example: “Which accountant for a small construction company in Quebec City?”
- Ask each one three times, in two or three assistants.Start a new conversation for every run and, if you can, sign out of your account.
- Write down what you see.Are you mentioned, and in how many runs? Is the description right: the correct service, the correct city? Which competitors keep coming up? When the assistant cites sources, which ones?
| Question | Mentions | Description | Competitors named |
|---|---|---|---|
| Accountant for a small construction company in Quebec City | 1 run of 3 | Correct | Firm A, Firm B |
| Accounting firm for a freelancer | 0 runs of 3 | Not applicable | Firm B |
| What does [your business] do? | 3 runs of 3 | Wrong city | None |
A mention in one run out of three doesn’t prove much, and a small change from one month to the next may only be noise. Repeated for three or four months with the same questions, the grid starts to show a direction, if there is one.
Two other sources fill in the picture. In Google Analytics, look at visits from chatgpt.com, perplexity.ai or gemini.google.com: some of that traffic arrives with no referrer and won’t appear in that report, but these are real visits. And add “an AI assistant” to the choices under “How did you hear about us?” on your forms.
What you can do
None of this guarantees a result. All of it also helps your search rankings and your customers, so it won’t be wasted if the assistants change tomorrow.
- Say plainly what you do.One page per service, spelling out what you offer, for whom and where. A phrase like “tailored solutions” gives nothing to quote. Also check that your robots.txt file isn’t blocking the crawlers you want to let in.
- Fix what’s out there.Google listing, industry directories, LinkedIn page, associations: same name, same address, same description. If an assistant puts you in the wrong city, that’s where the mistake may be.
- Answer the real questions.Service area, lead times, price range, types of customers: whatever people ask you on the phone deserves a written answer on your site.
- Earn mentions from others.Local press, partners, associations, customer reviews. It’s the slowest of the four, and assistants that read the web may draw on these sources.
What Oryon measures, and its limits
If redoing this grid by hand gets heavy, Oryon Index aims to automate part of the work: it looks at your presence on searches representative of your market, the themes linked to your brand, the sources it can observe and the competitors that appear.
It has the same limits as your spreadsheet. Its results are visibility signals rather than exhaustive measures of market share, it doesn’t know what goes on inside the models, and it can’t guarantee any mention. Read it the way you’d read your grid: over several months, looking for a direction.