Most businesses have no idea whether AI tools recommend them, because unlike traditional search there is no rankings report to check. You can fix that with a simple, repeatable measurement routine you run yourself. This guide gives you that routine step by step, using nothing more than the AI tools themselves and a spreadsheet.
The short version: pick a fixed list of prompts a real customer would use, run them across the major AI engines on a schedule, and record whether you appear, where, how you are described, and which sources the answer cites. The value is in the consistency and the trend, not any single answer.
Why can't I just check a dashboard
There is no universal AI visibility dashboard yet, because AI engines generate answers fresh each time rather than maintaining a public ranking index. A few paid monitoring tools have appeared, but the underlying method they use is the same one you can run manually: send prompts, capture answers, score them. Understanding the manual method first means you can interpret any tool you later adopt.
It also helps to know what you are measuring against. AI answers are not deterministic, so the same prompt can return slightly different wording each time. That is why a single check tells you little and a tracked routine over weeks tells you a lot. For the broader strategy context, see how AI search is changing local discovery.
What exactly should I measure
Measure four things for each prompt. Together they tell you not just whether you show up but how strong your position is.
- Presence. Were you named in the answer at all? This is the primary yes-or-no signal.
- Position. If named, were you first, second, or buried in a longer list? Earlier placement gets more attention, just like traditional rankings.
- Sentiment and description. How were you characterized? Positive, neutral, or with caveats? The adjectives reveal what the engine has learned about you.
- Citations. Which sources did the answer link to or reference? These show where the engine is getting its information and which pages to strengthen.
Recording all four turns a vague impression into a trend you can act on. If presence is low, you have a discovery problem. If presence is fine but sentiment is weak, you have a reputation problem, which our guide on reviews and reputation in AI recommendations addresses.
Step by step: build your prompt list
Your prompt list is the heart of the routine, so build it to match how real customers ask. Aim for ten to twenty prompts covering the genuine ways people describe their need.
- Start with category plus location. Best [your service] in [your city] and [your service] near [your city]. These are the workhorse prompts.
- Add problem-based prompts. Customers often describe a problem, not a category: who can fix [common problem] in [your city]. These reveal whether you get matched to needs, not just keywords.
- Add comparison prompts. What is the best [service provider] for [specific situation]. These surface whether you are recommended for your actual strengths.
- Add a few branded prompts. Tell me about [your business name] and is [your business name] any good. These show how the engine describes you when asked directly.
- Lock the list. Once set, keep the prompts stable so your month-to-month comparison is valid. Add new prompts over time, but do not edit the core set.
Step by step: run the routine
With your prompt list ready, run it the same way every time so the results are comparable.
- Pick your engines. At minimum, run every prompt through ChatGPT, Perplexity, and Google's AI features. These cover the major answer surfaces. Add others relevant to your audience.
- Use a clean session. Run prompts in a logged-out or fresh session where possible, so your own history does not bias the answer toward your business.
- Record into a spreadsheet. One row per prompt per engine per date, with columns for presence, position, sentiment, and cited sources. The structure is simple and the discipline is everything.
- Run on a schedule. Monthly is a sensible baseline. Run more often when you are actively making changes and want faster feedback.
- Note what changed. Alongside each run, jot any actions you took since the last run, such as new content or a batch of reviews, so you can connect cause to effect.
The whole pass takes an hour or two once your list is built. For where this sits in a full optimization program, see the answer engine optimization checklist.
What does a useful tracking spreadsheet look like
A useful spreadsheet is boring on purpose, with one consistent structure you fill in the same way every cycle. The discipline of a fixed format is what lets you spot trends, so resist the urge to redesign it each month.
A workable layout has these columns, with one row per prompt per engine per run:
| Column | What you record |
|---|---|
| Date | The run date |
| Engine | ChatGPT, Perplexity, or Google AI |
| Prompt | The exact prompt text |
| Present | Yes or no |
| Position | First, second, later, or none |
| Sentiment | Positive, neutral, or caveated |
| Sources cited | Which pages the answer referenced |
| Notes | Anything notable about the wording |
Keep a second tab for actions taken, dated, so you can line up changes against results. When presence improves two cycles after you launched a batch of question-answering pages, that connection is the payoff of keeping the records. Over a few months the spreadsheet becomes a quiet history of what moved the needle and what did not.
How do I read the results without fooling myself
Read the trend, not the snapshot, because AI answers vary run to run. A single absence is noise; a three-month decline in presence is a signal worth acting on.
A few interpretation rules keep you honest:
- Average across runs. Because answers vary, give a prompt a few runs in the same session and note the typical result rather than a lucky one.
- Separate discovery from reputation. Low presence points to discoverability work: consistent data, mentions, content. Strong presence with weak sentiment points to reputation work: reviews and how you are described.
- Follow the citations. If the same competitor's page keeps getting cited for a topic, that is the page to outdo with a clearer, more directly answerable one of your own. Our guide on how to get cited by ChatGPT covers the tactics.
- Compare engines. Strong on Perplexity but weak in Google's AI features is a meaningful pattern; our piece on Perplexity vs Google helps you decide where to focus first.
What do I do with what I learn
Turn each finding into one concrete action and re-measure next cycle. Measurement is only useful if it drives change, so close the loop every month.
The cause-and-effect map is straightforward. Missing entirely usually means a discovery gap, so tighten your business data and earn mentions. Present but described vaguely usually means thin reputation language, so improve your review flow. Present but for the wrong strengths usually means your content emphasizes the wrong things, so publish pages that answer the questions you want to win. As zero-click behavior grows, this kind of visibility matters more than ever, a shift we explore in the death of the click.
Measuring AI visibility is not complicated, but it does require the discipline to run the same routine consistently and read the trend rather than the snapshot. Once you can see where you stand, every improvement becomes testable. If you would rather have this measured and reported for you, request a free audit and we will benchmark your AI visibility across the major engines.
