Tracking AI search visibility: tools that actually work.

You track AI search visibility with a mix of purpose-built tools and manual prompt checks, because no single tool sees the whole picture yet. The category is young. The tools that exist do real work, but they measure fragments — one asks ChatGPT the same questions every week, another watches Perplexity citations, a third scrapes Google's AI Overviews. You stitch those fragments into something useful.

Most owners I talk to want one number: are we in the answers or not. That number does not exist yet, and anyone selling it to you is guessing. What you can get is a weekly read on whether the AI engines your clients use are naming you, quoting you, or linking you. That is enough to act on. Here is how to set it up without wasting money on dashboards you will never open.

What does AI search visibility actually measure?

AI search visibility measures whether AI engines mention, cite, or link your business when someone asks a question you should own. It is not the same as a keyword ranking. A ranking is a position on a page of blue links. AI visibility is a presence inside a generated answer — sometimes as a named recommendation, sometimes as a source link, sometimes as a quoted line of text.

There are three things worth counting. First, mentions — does the engine say your name. Second, citations — does it link your site as a source. Third, sentiment — when it names you, does it say something accurate and useful. A San Diego family law firm might get named in a ChatGPT answer about filing timelines but described with a fee range from a competitor's page. That is a mention with bad data attached. You want to catch that.

The engines that matter for most SMEs are ChatGPT, Google's AI Overviews, and Perplexity. Gemini and Claude matter less for local commercial queries right now, though that shifts. Pick the two or three your clients actually use before you buy any tool.

Which tools actually track AI visibility?

The tools that work fall into three groups: dedicated AI-rank trackers, citation monitors, and manual prompt logs. Each covers a gap the others leave open. You do not need all three. You need one paid tool and one manual habit.

Dedicated AI-rank trackers run a fixed set of prompts against ChatGPT, Perplexity, and Gemini on a schedule and record whether you appear. Profound, Peec, and Otterly.ai sit here. They cost real money — Profound and Peec are built for teams and price accordingly. Otterly.ai is the one I point solo operators toward because the entry tier is affordable and the interface does not assume you have an analyst on staff. You give it your prompts, it checks them weekly, it shows you which engines named you and which named a competitor.

Citation monitors watch for your domain appearing as a source link inside AI answers, mostly Perplexity and AI Overviews. Some SEO suites — Ahrefs and Semrush both shipped AI-overview tracking in the last year — now flag when your pages get pulled into Google's generated answers. If you already pay for one of those, turn that feature on before you buy anything new. You may already have half of what you need.

Manual prompt logs are a spreadsheet. Ten questions your best clients ask, run by hand across ChatGPT and Perplexity once a month, results pasted into a doc with the date. This sounds primitive. It is the most honest signal you will get, because you read the full answer in context instead of a green tick in a dashboard. I still do this for our own practice and for clients in the first ninety days.

How do you set up tracking without wasting money?

You set up tracking by writing your prompts first, choosing one paid tool second, and keeping a manual log third. The order matters. Most people buy the tool before they know what to measure, then track prompts that no real client would ever type.

Start with the prompts. Write down the ten questions a good client asks before they hire someone in your field. Not keywords — questions. "How long does an uncontested divorce take in California." "What does a roof inspection cost in San Diego." "Do I need a lawyer for a small estate." These are the prompts you feed the tools. If your business does not appear for these, ranking for a vanity term will not save you.

Then pick one paid tool. If you run a solo or small-group practice, Otterly.ai or the AI feature inside an SEO suite you already pay for. If you have someone handling marketing part-time, Peec gives them more to work with. Do not run three trackers at once. You will not read three dashboards, and the numbers will disagree, and the disagreement will paralyze you.

Then build the manual log. First of each month, run your ten prompts by hand. Record: engine, did we appear, were we cited with a link, was the information correct. Fifteen minutes. Over six months that log tells you more than any tool, because you can see the wording change and catch the moment an engine starts describing you accurately.

Here is the checklist I hand clients:

1. Write ten real client questions.
2. Pick two engines your clients use.
3. Choose one paid tool or turn on the AI feature in your existing SEO suite.
4. Run all ten prompts manually today and save the answers as your baseline.
5. Repeat the manual run monthly. Read the tool weekly.
6. Fix the source pages when the data is wrong, not the prompt.

What do the numbers actually tell you?

The numbers tell you which pages the AI engines trust enough to quote, and that points you straight at your foundation. When an engine cites a page, it is telling you that page is clean, fast, and clear enough to pull from. When it skips you and quotes a competitor, the fix is almost never a new tool. It is the page.

AI engines pull from pages they can read quickly and parse cleanly. A page that loads slowly, buries the answer under a hero image, or hides the real content behind a script gets skipped. This is the same discipline that governs whether Google bothers to rank you in the first place — we wrote about that in Core Web Vitals: the three numbers that decide if Google bothers. The engines reward the same things: speed, structure, a direct answer near the top.

So read your tracking data as a page audit in disguise. If Perplexity cites your competitor's FAQ page for a question you answer better, go look at your version. Is the answer in the first sentence, or three paragraphs down. Is the page fast. Does it name the specific thing — city, cost, timeline — or does it hedge. The engines quote specifics. They skip mush.

This is why speed keeps showing up in AI visibility work even though it feels like a technical concern. A slow page loses the crawl, loses the citation, and loses the client who bounced before it loaded — which is the argument we made in why your slow site is a sales problem, not an IT problem. The tracking tools surface the symptom. The Search Foundations work fixes the cause.

We saw this play out with McShanes Solicitors. Cleaning up the foundation — fast pages, clear answers, structured content — moved them into positions the engines could actually read. The tracking confirmed it. It did not create it.

Where does AI visibility tracking break down?

AI visibility tracking breaks down because the answers are not stable and no tool sees every engine. Run the same prompt twice in ChatGPT and you can get two different answers with two different sources. The engines personalize, they update models, and they change how they cite without warning. A dashboard showing you at 80% visibility on Tuesday can show 60% on Thursday with nothing changed on your end.

That volatility means you should read trends, not days. One week of data is noise. Six weeks of data is a signal. Do not let a bad Tuesday send you rewriting pages that were working fine.

The second limit is coverage. No tool tracks every engine well, and the engines people actually use shift by industry and age. Some tools are strong on Perplexity and weak on AI Overviews. Others reverse that. This is why the manual log stays in the mix — it catches what the tool misses and keeps you honest about which engines your clients truly use.

What we won't tell you

We won't tell you AI visibility is a solved measurement problem, because it is not. The tools are early, the numbers move, and anyone promising a clean single score is selling confidence they cannot back. We also won't tell you to chase every engine — for most SMEs, two engines and ten real questions cover the ground that matters. Pick a tool, keep a manual log, and read the trend over months. The rest is patience.

Forget the tools for a second. The work underneath is the same work that has always mattered: pages that load fast, answer the question, and tell the truth in specifics. Track what you can, fix the foundation, and let the numbers confirm the work rather than replace it. Foundations first.