Most AI visibility dashboards are built to make you feel good, not to make you money.
I say that as someone who builds one. I run LLM Pulse, an AI search visibility platform, so treat everything here with the healthy suspicion you would apply to any founder writing about his own category. My promise is simple: I am going to tell you which metrics I would ignore even if my own product showed them off, and which ones I would fight to keep even if they made the chart look worse.
Because here is the uncomfortable truth about this whole space. We spent two years teaching brands to obsess over whether ChatGPT "mentions" them. And a mention, on its own, is worth almost nothing.
The vanity trap
A vanity metric is one that goes up reliably, feels like progress, and does not change a single decision you make.
In AI visibility, the classic offenders are raw mention counts and "we appeared in an AI answer" screenshots. You track how many times a model said your name across a batch of prompts, the number climbs, everyone claps.
The problem is that raw counts hide everything that matters. A mention where the model calls you "a cheaper but less reliable option" counts exactly the same as one where it calls you the best choice for enterprise teams. A mention buried in a 900-word answer nobody finishes counts the same as one in the first sentence. And a mention on a prompt no human ever types counts the same as one on the prompt that drives your pipeline.
Raw mention volume is the AI-era equivalent of impressions. Interesting context, terrible north star.
So let me go through the metrics that survive scrutiny, roughly in the order I would build a reporting stack around them.
Presence rate, not mention count
Start with presence rate: across your tracked set of prompts, in what percentage does your brand appear at all?
The shift from a count to a rate matters more than it looks. A count of 340 means nothing until I know the denominator. A presence rate of 62 percent across 150 commercial prompts tells me exactly where I stand and gives me a clean number to move.
The discipline is in the prompt set. If you track 150 prompts your customers would actually ask and your presence rate is 62 percent, that is a real baseline. If you cherry-pick 30 prompts where you already win, you have built a mirror, not a measurement. Choose prompts by buyer intent, keep the set stable, and only then does the rate mean anything month over month.
One honest caveat: models are non-deterministic. Ask the same question twice and you can get different answers. Any tool worth paying for samples each prompt multiple times and reports a rate, not a single lucky roll. If your vendor shows you one answer per prompt and calls it data, that is a red flag.
Share of voice, because you exist next to competitors
Share of voice is presence rate with your competitors in the frame. Of all the brand mentions in answers to your prompts, what slice is yours?
This is the metric I trust most for reading the actual market, because it is relative. You can have a great month, push your presence rate up eight points, and still lose ground because two competitors moved faster. Absolute presence flatters you. Share of voice keeps you honest.
It is also the metric that most naturally maps to how buyers experience AI answers. Nobody sees your brand in isolation. They see a shortlist the model assembled, and your job is to be on it and near the top of it. Share of voice measures exactly that.
Track it per model and per market. You will often be strong in Perplexity and weak in Google AI Overviews, or dominant in your home country and invisible one border over. An aggregate number hides those gaps, and the gaps are where the work is.
Sentiment and framing, because being mentioned wrong is worse than not being mentioned
Here is where counts fall apart completely.
Sentiment is what the model actually says about you. Not whether you appear, but how. Are you the recommended option or the cautionary footnote? Does it repeat a strength you are proud of, or a criticism from a two-year-old review that is no longer true?
This one is underrated and it is where real money leaks. I have seen brands celebrate rising mention counts while the models were quietly describing them as expensive, or hard to set up, or worse, confusing them with a competitor and attributing someone else's flaws to them. Every one of those mentions was a small loss dressed up as a win.
What you want is the exact wording, per model, over time, with the negative or wrong framing flagged so you can go fix the sources feeding it. A number from one to five is a start. The sentence the model wrote is the actual work. In LLM Pulse we surface the literal language and flag the negative or inaccurate framing, because "sentiment: 3.4" tells you nothing about what to change.
Citations and sources, because they are your only real lever
Mentions and sentiment tell you where you stand. Citations tell you why, and they are the closest thing to an action list this field offers.
When a model answers with sources, those cited pages and domains are what shaped the answer. If a comparison site, a Reddit thread, and a competitor's own blog are the three sources behind an answer where you lose share of voice, you now know precisely what to influence. You cannot edit the model. You can influence what it reads.
So track which domains and pages get cited for your prompts, how often, and whether you are among them. This is the metric that turns a visibility report from a scoreboard into a plan. Everything upstream is diagnosis. Citations are the prescription.
The metric that outranks all of the above: real traffic and conversions
Now the argument I actually came here to make.
Every metric above lives inside the AI answer. Presence, share of voice, sentiment, citations. All of it measures what happens on the model's screen. None of it, by itself, proves a single human left that answer, came to your site, and did something worth money.
If you cannot connect AI visibility to traffic and revenue, you are optimising a proxy.
This is the gap I care most about, and it is the one most of the category quietly steps around, because measuring answers is easy and measuring outcomes is hard. But the outcome is the entire point. A 70 percent share of voice that sends no traffic and closes no deals is a lovely slide and a waste of a quarter.
The good news is that the AI referral trail is finally traceable. When someone clicks through from ChatGPT, Perplexity, or Gemini, that visit lands in your analytics with an identifiable source. Connect your analytics to your visibility data and you can follow the chain end to end: this prompt, this model, this citation, this many visits, this many conversions, this much revenue.
That is the connection LLM Pulse is built around. You link GA4 or Plausible and see the actual visits, conversions, and sales that ChatGPT, Perplexity, and Gemini send you, sitting right next to the mentions. Not "you appear 62 percent of the time" in one tool and "something changed in organic" in another, with you squinting to connect them. The mention and the money in the same view.
Once you can see that, your whole prioritisation flips. You stop chasing share of voice on prompts that convert nobody and start defending the ten prompts that quietly send buyers. Some of your proudest visibility wins turn out to be dead traffic. Some unglamorous prompt you almost dropped turns out to be your best channel. You only learn that by tying mentions to outcomes.
One more you should watch: bot and crawler access
A quieter operational metric: are the AI crawlers actually reaching your pages?
If GPTBot, ClaudeBot, or PerplexityBot cannot fetch a page, it cannot become a citation, full stop. Tracking crawler access per page is unglamorous plumbing, but it is often the difference between a page that earns citations and one the models never see. Worth a panel in your stack, not a headline.
How I would rank them
If I had to put it on one line, in priority order:
- Real traffic and conversions from AI
- Share of voice on high-intent prompts
- Citations and sources you can influence
- Sentiment and framing
- Presence rate
- Raw mention count (context only, never a goal)
Notice the thing everyone leads with, the mention count, is dead last.
A note on choosing a tool, honestly
Because I build one of these, I owe you a fair map of the rest, and the honest version is that the best tool depends on who you are.
If you are a Fortune 500 brand that needs the deepest data, prompt volume estimates, procurement-grade security, and a real API, Profound is the serious pick and I will not pretend otherwise. It is enterprise, it is demo-led, and it is priced for that world. If you are a team or agency that wants strong regional and multi-language tracking with clean per-country competitor comparison, Peec AI is genuinely good, just note that models beyond the included three are paid add-ons priced per model. If you want the cheapest honest way to start and to eyeball real AI-output screenshots, Otterly.ai starts at 29 dollars a month. And if you already pay for Ahrefs or Semrush, their AI-visibility features are shallower than the specialists but they are already in your subscription, which for many teams is the right trade.
Where I would put LLM Pulse is exactly on the argument of this article. We start at 49 €/month with public pricing, unlimited seats on every plan, and a 14-day trial, and the thing I would actually pick us for is closing the loop between mentions and money: GA4 and Plausible connected to real AI traffic, sentiment in the model's own words, citation tracking, five core models on every plan without an enterprise call, and a hosted MCP server so you can pull your data straight into Claude or ChatGPT. For agencies, the full white-label (your logo, your domain, a separate project per client) is a real reason to look.
Here is the pricing as of July 2026. Prices change, so check before buying.
| Plan | Price (per month) | Prompts | Notable additions |
|---|---|---|---|
| Starter | 49 € | 50 | 5 competitors, 5 core models |
| Growth | 99 € | 150 | Sentiment, AI traffic analytics, agent |
| Scale | 299 € | 450 | Reputation, GEO testing, ChatGPT Shopping, REST API, Data Studio connector |
| Higher tiers | 543 €, 1.086 €, then custom | More | Enterprise volume |
Every plan includes unlimited seats and the hosted MCP server. That last part is still rare in this category.
The line I want you to remember
An AI mention you cannot connect to a visitor is a rumour. Measure the ones that show up in your bank account.