You cannot improve what you do not measure, and right now most businesses have no idea what AI assistants say about them. Ask ChatGPT or Gemini to recommend a business in your market and you will get names. Whether yours is among them, how you are described, and whether your website is credited as a source: all of that is knowable, and hardly anyone is checking.
The good news is that tracking AI visibility needs no special software and no big budget. A spreadsheet, a fixed set of questions and an honest hour each month will tell you more about where you stand than most agencies could have told you a year ago. This guide walks through exactly what to measure, how often, and how to keep the measurement honest.
If the field itself is new to you, our plain-English guide to answer engine optimisation covers the why. This one is pure practice: the questions, the routine and the signals that move the numbers.
Build the question set your buyers actually use
Start by writing down ten to fifteen questions a real buyer in your market would put to an assistant. Mix three kinds: direct recommendations (“who is the best solicitor in Galway for buying a house?”), cost and process questions (“how much does a house survey cost in Ireland?”) and comparisons (“is a local firm or a national one better for this?”). Write them as full sentences, because that is how people talk to these tools; nobody asks Claude “solicitor Galway conveyancing”.
Then freeze the list. The value of tracking comes from asking the same questions over time, not from finding the perfect questions. Add new ones at the bottom if your services change, but never quietly retire a question because its answers disappoint you. That is how you cheat yourself out of the truth.
Run the set through each assistant, monthly
Once a month, put every question to ChatGPT, Claude, Gemini and Perplexity. Use a fresh conversation for each question so earlier chat history does not colour the answer, and do the run on roughly the same day each month so the gaps between snapshots stay even.
Expect variance. The same question can produce different answers an hour apart, because these systems compose each reply fresh, usually running their own web searches and reading whatever ranks that day, rather than reading from a fixed list. That is normal, and it is exactly why the monthly trend matters far more than any single answer. One flattering mention proves very little; a rising count of mentions across four assistants and a dozen questions is a real signal you can act on.
Record three things for every answer
Keep the spreadsheet simple: one row per question, one column per assistant, one tab per month. For every answer, record three things:
- Mentions: was your business named at all, and how prominently did it appear in the answer?
- Citations: was your website linked or credited as one of the sources the assistant drew on?
- Description: what did the tool actually say about you, and was it accurate and current?
Record the competitors who get named too. Over a few months you will see who owns each question in your market, which can be uncomfortable reading, and that is precisely the point. The description column earns its keep as well: an assistant confidently describing services you no longer offer, or quoting an old address, is a fixable problem you would otherwise never know about.
The signals that move the numbers
When the mentions are not coming, resist the urge to chase tricks; there is no secret prompt and no loophole. The engines reward a short list of honest signals. An assistant cannot check whether your claims are true; it weighs whether the signals around your business agree, and reads that agreement as trust, so the whole job is making your real signals line up. A consistent identity, meaning your name, services and location described the same way on your site, your profiles and every directory that lists you. Plain answer content: pages that address your buyers’ real questions in clear language an assistant can lift and credit accurately. Trusted mentions, earned through genuine link building and digital PR, because assistants lean on sources the wider web already vouches for. And schema markup, the structured data layer of solid technical SEO, which spells out who you are in a format machines can read without guessing.
None of this is exotic, and every bit of it strengthens your Google rankings at the same time. That overlap is the quiet advantage of measuring AI visibility: the fixes it points you towards are rarely wasted anywhere else. We cover that relationship properly in GEO vs SEO: what’s the difference?
Keep the measurement honest
Two warnings before you build the spreadsheet. First, measurement does not make the outcome controllable: nobody can guarantee a mention in a ChatGPT or Gemini answer, and no tracking routine changes that. What tracking gives you is the truth about your trend, which is worth more than any promise anyone could make you. Second, record every answer, including the deflating ones. A log that only keeps the flattering responses is marketing, not measurement, and it will steer your decisions wrong.
Two ways to get your first baseline
If you would like a feel for where you stand before committing to the routine, our AI visibility checker is a free two-minute self-check against these same signals, no email address required. The real scan lives inside the free SEO audit: we run your market’s questions across ChatGPT, Claude, Gemini and Perplexity and hand you the recorded results alongside your rankings and site health. And if you would rather the monthly routine ran itself, our AI visibility and GEO service does this exact measurement for you, reported in plain English, with the trend shown honestly whether it flatters us or not.
