Is Ahrefs Brand Radar legit or is AI visibility tracking still mostly vibes?

The Problem: Everyone Sells AI Visibility Scores, Nobody Agrees

Ask three AI visibility vendors to measure your brand's presence in ChatGPT, and you'll likely get three different numbers. That's not bad luck — it's the current state of a category where more than 20 companies sell AI visibility measurement tools using divergent methodologies that, per the IAB, can produce different answers for the same brand.

The IAB's assessment is blunt. Without a shared evaluation standard, says Caroline Giegerich, VP of AI at the IAB, "the industry can't tell reliable signals from noise." That's why the organization introduced a two-tier framework separating merely "directional" data from "decision-grade" data — and why directional scores alone aren't sufficient for budget allocation.

Then there's the deeper technical problem: LLMs are non-deterministic. The same prompt run on the same engine at the same time can return different responses. How much does that matter in practice? A lot, according to a SparkToro study of 2,961 runs across ChatGPT and Google AI: the same brand list appeared in the same order only about 1 time in 1,000.

Zapier's hands-on review of roughly 50 tools reached a similar conclusion: "It's more art than science at this point, so you should expect some fluctuation in the results." If a single run can flip your brand's position, then a one-time check — "we asked ChatGPT and we showed up!" — is anecdote, not measurement.

• Small, single-LLM prompt sets produce a data sample, not proper measurement — if the prompts don't reflect real customer questions, you're paying for noise.
• Referral-based measurement undercounts badly: over 97% of brand site visits following a ChatGPT mention carried no UTM parameter, per one analysis of more than 2 million conversations.
• No single tool checks every box yet — the category is simply too new for that.

The practical takeaway: visibility percentage across many runs is worth tracking; a one-time mention is not. Repeated sampling, triangulated with branded search trends and actual revenue, turns vibes into signal. The same principle applies beyond visibility dashboards — when evaluating any AI capability, from a tracking tool to the agent skills your team installs, repeated testing beats a single impression. It's the mindset behind how SkillzBench approaches agent skills: benchmark each one in a sandbox, with and without the skill, rather than trusting a demo.

Until vendors converge on the IAB's decision-grade criteria — query volume, sample size, prompt coverage, testing cadence, reproducibility, and platform coverage — treat every single-check score as a starting hypothesis, not a fact.

What Brand Radar Actually Does (Straight from Ahrefs' Own Methodology)

Most AI visibility tools ask you to take their numbers on faith. Ahrefs Brand Radar is the rare exception — it publishes its methodology, including the parts that make its own data look worse.

That transparency starts with the core admission: Brand Radar's metrics are "directional indicators, not exact traffic counts," per Ahrefs' own methodology post. Estimated Impressions are weighted by Google search volume, which Ahrefs describes as "a modeling choice, not a measured relationship." Even hallucinated or malformed links are deliberately left in, because they "reflect real model output."

The scale behind the tool is real. Brand Radar draws on 110 billion discovered keywords (28.7 billion with positive search volume) and runs monthly query volumes across major engines — roughly 14.5M for ChatGPT, 14.6M for Perplexity, 14.6M for Gemini, and 13.3M for Copilot — with a 90-day reporting window. It's also the only tool in independent roundups tracking social platforms like YouTube, Reddit, and TikTok alongside AI engines.

What you pay for it:

• $199/month as an add-on, with no free trial
• Custom prompts starting at $50
• Cheaper alternatives exist — OtterlyAI from $29/mo, ZipTie from $35.63/mo
• Coverage is strongest in English; non-English markets are only proportionally represented

Independent reviewers land on "solid, not category-leading." Zapier, after testing roughly 50 tools, lists Brand Radar only as "best for existing Ahrefs users," while Brand24 names it "best for competitor benchmarking" — its author personally uses it to report AI visibility to stakeholders.

The honest caveat: even Ahrefs positions the tool for benchmarking AI Share of Voice and competitor comparison, explicitly calling it "not a substitute for audience measurement or traffic analytics." That matters because the broader category is still immature — the IAB reports 20+ vendors using divergent methodologies, and its framework says directional data isn't sufficient for budget allocation. LLM non-determinism compounds this: a SparkToro study of 2,961 runs found the same brand list returned in the same order only about 1 time in 1,000.

The practical takeaway is triangulation. Since research on AI-driven visits shows over 97% of brand site visits after a ChatGPT mention carry no UTM parameter, visibility tools and traffic analytics measure different things. Pair Brand Radar's trend lines with branded search data and CRM revenue rather than treating any single score as ground truth.

That same skepticism — verify claims against real output rather than trusting vendor promises — is exactly why SkillzBench is building free sandbox benchmarking for AI agent skills, testing each one with the same model with and without the skill and publishing the measured lift. Whether you're tracking brand visibility or evaluating a skill, repeated, triangulated measurement beats a single impressive-sounding number every time.

The Verdict: Directional Yes, Decision-Grade No

So, is Brand Radar legit? Yes — and Ahrefs itself will tell you exactly how far that "yes" goes. The honest answer lives in the gap between two words the IAB uses to grade measurement quality: directional and decision-grade.

The IAB's framework sets six criteria for decision-grade data: query volume, sample size, prompt type coverage, testing cadence, reproducibility, and platform coverage. Its position is unambiguous — directional data is not sufficient for budget allocation. And Ahrefs' own methodology documentation labels Brand Radar's metrics "directional indicators, not exact traffic counts," built on a modeling choice (weighting impressions by Google search volume) rather than a measured relationship. That's a rare level of candor, but it means the tool fails the decision-grade bar by its own admission.

Reproducibility is the deepest problem, and it isn't Ahrefs-specific. A SparkToro study of 2,961 runs found the same brand list returned in the same order only roughly 1 time in 1,000 — LLMs are non-deterministic by nature. No vendor can fully solve that today.

Then there's the referral blind spot, which quietly distorts any attempt to validate visibility data against your analytics:

• Over 97% of brand site visits following a ChatGPT mention carry no UTM parameter, so they vanish into "direct" traffic (Profound's study of 2M+ US conversations).
• Bing Chat click-through rates run 51–94% lower than Bing web search, per Microsoft data from a court filing.
• AI-driven visits are real despite the tracking gap — Profound found 1.5–2.5x baseline visit lift within 7 days of an AI mention.

The practical takeaway: use Brand Radar for benchmarking, trends, and competitor comparison — never as the sole basis for spend decisions. Triangulate across three layers, as Arcalea recommends: exposure (visibility scores), behavior (branded search and direct traffic trends), and outcome (CRM revenue). Referral analytics alone give you a floor, not the full picture.

If you're operationalizing this workflow — scheduled visibility checks, prompt-set testing, cross-tool reporting — agent skills can automate much of the plumbing. SkillzBench's free skills directory and MCP connector are being built for exactly that: discover, verify, and equip skills so your agents handle the repetitive measurement work while you focus on interpretation.

Single checks are vibes; repeated, triangulated measurement is signal. That's the standard to hold every AI visibility vendor to — including this one.

How to Use AI Visibility Data Without Getting Fooled

A single ChatGPT check tells you almost nothing. A SparkToro study of 2,961 runs found the same brand list returned in the same order only about 1 time in 1,000 runs — so any visibility score based on one query is noise, not signal.

The practical playbook starts with repetition. Track visibility percentage across many runs, never one-time mentions. As Brand24's Marta Zwierz puts it, if the prompt set is small, single-LLM, or doesn't include the real questions your customers ask, "you're paying for a data sample, not proper data measurement."

Next, triangulate. Referral analytics undercount AI badly: over 97% of brand site visits following a ChatGPT mention carried no UTM parameter, according to Arcalea's analysis. Their framework layers three measurements:

• Exposure — mentions and citations across AI engines, tracked over many runs.
• Behavior — branded search volume and direct traffic trends, which capture AI-driven visits referral data misses.
• Outcome — revenue attributed in your CRM, the only layer that justifies budget.

Then hold any vendor to a standard. The IAB's framework defines six decision-grade criteria: query volume, sample size, prompt type coverage, testing cadence, reproducibility, and platform coverage. Ask each vendor directly how they score on all six — and whether their prompts reflect what real customers actually type.

Budget matters too. Brand Radar costs $199/month as an add-on, while OtterlyAI starts at $29/month and Peec AI at $95/month. For basic tracking, the cheaper tools may be enough. Brand Radar earns its price in one niche: it's the only tool tracking social platforms — YouTube, Reddit, TikTok — alongside AI engines, which matters if your buyers research there. Ahrefs also runs prompts through the free public web interfaces of the AI platforms, which better reflects what actual users see.

Whatever you pick, keep expectations calibrated. Ahrefs itself calls Brand Radar's metrics "directional indicators, not exact traffic counts," and the IAB says directional data isn't sufficient for budget allocation. Use these tools to benchmark trends and competitor share of voice — then let behavioral and revenue data make the spend decision.

One final note on rigor: the same skepticism applies to the AI tools you deploy, not just the ones you measure with. Published research found roughly 26% of community-published agent skills carry at least one vulnerability, which is why SkillzBench is building free security scanning, sandbox benchmarking, and version-pinned verification for agent skills. Treat every AI-derived number — and every AI tool — as something to verify, not vibes to trust.

From Measurement to Action: Automating the Follow-Through

Knowing where AI mentions you — and where it doesn't — only matters if something happens next. The practical follow-through is unglamorous: scheduled visibility checks across many runs, citation-gap reports showing which sources AI engines cite that you don't appear in, and content updates targeting the prompts where you're absent. Because LLMs are non-deterministic, a hands-on reviewer who tested roughly 50 tools warns results will fluctuate and calls the discipline "more art than science" — a SparkToro study of 2,961 runs found the same brand list in the same order only about once in 1,000. Single checks are noise; repeated checks are signal.

That repetition is exactly where manual workflows break down. Brand24's reviewer puts it plainly: visibility percentage across many runs is worth tracking, while a one-time mention in ChatGPT is not. But running the same prompt set weekly across ChatGPT, Perplexity, and Gemini, logging changes, and turning absences into content briefs is hours of work nobody reliably does. It's also where errors creep in — inconsistent prompt wording, forgotten engines, gaps in the log.

• Schedule recurring visibility checks so the same prompt set runs on a fixed cadence, not whenever someone remembers.
• Generate citation-gap reports comparing the sources AI engines cite against the pages you actually publish.
• Trigger content updates when visibility drops on prompts that reflect real customer questions, not generic queries.
• Triangulate findings with branded search trends and CRM revenue, since over 97% of visits after a ChatGPT mention carry no UTM parameter and referral analytics undercount AI impact.

This is where agent skills earn their keep. A well-built skill can package the entire workflow — run the prompt set, compare against last check, flag citation gaps, and draft the update brief — as repeatable instructions any coding agent or chat-app agent can execute. Instead of a person re-running Brand Radar-style checks by hand, the agent does it on schedule and the team reviews the output.

The catch, given that published research found roughly 26% of agent skills carry at least one vulnerability, is trusting which skills are safe to wire into your workflow. That's the problem SkillzBench is building toward: a free directory of validated, security-scanned agent skills where every version is pinned to a commit SHA, every skill is benchmarked in a sandbox with and without the skill, and each carries a 0–100 trust score — plus a free MCP connector so any agent can run these visibility checks with only the skills you've chosen. The waitlist is live for teams that want this follow-through automated from day one.

Measurement without follow-through is just expensive dashboard-watching. Whether you check manually or hand the cadence to an agent, the principle holds: many runs, fixed prompts, documented gaps, and content fixes — repeated until the absence becomes presence.

ctaText: Browse and test free agent skills on SkillzBench socialProofText: SkillzBench is building a free directory of security-scanned, sandbox-benchmarked agent skills — every version pinned, trust-scored, and shown with its measured Lift separately. Join the waitlist.