$300M Went Into AI Visibility Tools in a Year. Here's How a Lean B2B Team Should Actually Pick One.

Between summer 2025 and spring 2026, tools that track whether AI assistants mention your brand raised more than $300 million.^[1]

That number is the story. Not because funding proves a category works, but because it explains what a buyer now walks into. Eighteen months ago there were a handful of scrappy prompt trackers. Today there is a tiered market with enterprise security certifications, European challengers, and an SEO-suite tab in every tool you already pay for.

The category matured faster than the measurement discipline underneath it. Which is why the buying decision is no longer "should we track AI visibility." It is "what actually distinguishes these tools, and which of those differences do we need."

What the funding tells you about the tiers

Profound went from a $3.5M seed in August 2024 to a $20M Series A from Kleiner Perkins, a $35M Series B from Sequoia, and a $96M Series C from Lightspeed at roughly a $1B valuation, about eighteen months after launch. It tracks 10+ AI engines and carries SOC 2 Type II.^[2]

Berlin-based Peec AI raised a $21M Series A in November 2025, reports $4M+ ARR, and says it tracks 1,300+ brands across up to ten AI models including ChatGPT, Perplexity, Google AI Overviews, Gemini, Google AI Mode, Claude, DeepSeek, Copilot, Grok and Llama.^[1] For DACH buyers, the European base is not a patriotic detail. It is a data residency conversation you will otherwise have with your own legal team six weeks into procurement.

Evertune has raised $19M from adtech and martech investors.^[3]

And Semrush and Ahrefs have both added AI monitoring, though as additions layered onto existing SEO infrastructure rather than purpose-built answer-engine platforms.^[2]

Read that as three tiers, because that is the actual choice on the table.

Purpose-built AEO platforms. Deepest engine coverage, the most developed methodology, the highest price. Worth it when AI visibility is a named objective with a budget line and someone accountable for it.

SEO-suite bolt-ons. Cheapest to adopt because you already have the contract and the team already logs in. Coverage is narrower and the methodology usually less transparent. Genuinely sufficient when you want a directional read alongside your rankings, and you are not going to staff dedicated work against it.

DIY prompt harnesses. A scheduled script that runs your prompt set against the APIs and writes results to a sheet or warehouse. Underrated for lean teams. You own the data completely, you control the sampling, and it costs API credits. It gives you no benchmarking and no interface anyone else on the team will open.

Most lean B2B teams are honestly in tier two or three and get sold tier one.

A warning about the rankings you are reading

Search "best AI visibility tools" and you will find ranked lists with scores. Check the domain on each one. A large share are published by vendors inside the category, by agencies selling GEO retainers, or by affiliates. The comparison is the marketing.

That does not make the facts in them wrong. It does mean the ranking order is not evidence, and the "AEO score" in the headline is the publisher's own construct.

There is a second comparability problem that has nothing to do with bias. AI visibility scores are sample-dependent. Two tools running different prompt sets, different numbers of runs per prompt, different locales and different engines will produce different numbers for the same brand in the same week, and neither is wrong. A score from tool A and a score from tool B are not the same unit. We covered the underlying sampling math separately in our piece on how many prompts you should actually track, and the short version is that most dashboards report a movement long before it is distinguishable from noise.

The evaluation checklist

This is the part to bring into the procurement process.

1. Engine coverage, and how it is obtained. Ask whether the vendor runs real queries against the assistant interfaces or infers results from underlying model APIs. These produce different answers, because the consumer product wraps retrieval, grounding and safety layers the raw API does not. A tool that queries the API and calls it "ChatGPT visibility" is measuring something adjacent to what your buyer sees.

2. Prompt-set ownership. Can you export your prompts, your raw responses and your historical results, in full, at any time? If the answer is a dashboard screenshot, your entire history is hostage to the subscription.

3. Disclosed sampling methodology. Runs per prompt. Sampling schedule. Whether results are deduplicated. Whether the tool reports any measure of variance. A vendor that cannot tell you how many times it ran each prompt is selling you a number without a denominator.

4. Locale and region. Most tools default to US English. If you sell into DACH, you need German-language prompts, German locale signals, and results that reflect what a Munich buyer sees. Ask for a live demo query in German, on your own brand, during the call. The gap between the marketing claim and the demo is frequently visible in thirty seconds.

5. Data residency and hosting. Where do prompts and responses live, and under which jurisdiction? This is where the European vendors have a structural advantage and where the question is cheaper to ask before the contract than after.

6. Why, not only whether. The useful question is not "were we mentioned." It is "which sources did the model draw on for that answer, and were we in them." A tool that shows cited domains and source attribution tells you where to work. A tool that shows only a mention percentage gives you a metric and no next action.

7. Connection to action. Does the tool hand off into content and citation-source work, or does it terminate in a report? Reporting-only tools quietly become the thing nobody opens in month four.

8. Pricing as the prompt set grows. Price per prompt, per brand, per engine, per run. Model it at three times your starting prompt count, because that is where you will be once the team gets interested.

9. Export and API. If the data cannot land in your warehouse next to pipeline, it will never be in a board deck.

Questions to ask on the demo call

Literally these, in this order.

• How many times do you run each prompt, and over what window?
• Do you query the consumer assistant or the model API?
• Can I export every raw response, not just aggregates?
• Show me a German-language query on my brand, right now.
• Where is the data stored and processed?
• When my visibility drops, what exactly does your product tell me to do?
• What does this cost when my prompt set goes from 50 to 200?
• Which of the numbers on this dashboard would move if I changed nothing?

That last one is the most informative question in the list.

What the tooling will not do for you

The tool measures. It does not earn.

The research on what actually shifts generative-engine visibility is less exciting than the dashboards imply. A 2026 critical survey of 45 studies concluded that generative engine optimization is not one ranking trick but a multi-stage pipeline, and that very few popular tactics survive scrutiny.^[4] The lifts most often quoted from the original GEO research were +41% from adding quotations, +32% from statistics, +30% from citations and +28% from fluency optimization.^[5] Those are content-structure interventions, not settings in a platform.

And much of the citation work happens off your domain entirely, in third-party sources the model trusts. No tracker changes that. It only tells you it happened.

The reason to measure anyway is that the traffic behaves unusually well. Vendor-aggregated figures put LLM visitor conversion at roughly 15.9% from ChatGPT, 10.5% from Perplexity and 5% from Claude against about 1.76% for organic search, with AI-referred sessions up 527% year over year across the first five months of 2025.^[5] Those come from vendor aggregations with mixed methodologies, so treat the magnitude as directional. The direction has been consistent across every dataset we have seen.

The actual decision

Buy the cheapest tool that answers "why were we not cited" in your language and your market, and that lets you export everything.

Then spend the difference on the content and source work the tool tells you to do. A team with a tier-two tracker and a working content loop will out-rank a team with a tier-one dashboard and no one assigned to act on it. That is not a controversial claim. It is just the one that vendor comparisons cannot make.

Nukipa is built for the second half of that sentence: the loop that turns a visibility signal into published, current, cited content, with every decision traceable to a reason you could say out loud. If you want to see what your measurement looks like when it is wired to the system that acts on it, test Nukipa.