See exactly when AI recommends you.
WhyIQ AI Radar measures your AI search visibility. Every week it runs your real buyer-intent prompts through ChatGPT, Perplexity, Claude, Gemini, and Google's AI Mode, then reads back exactly which sources each engine cited. Not a prediction: a measurement of what the engines actually said.
Below is the actual product view, live from the app with sample data. This is what lands in your inbox every week.
AI engines regularly recommend Fernway Coffee.
First source cited, across 128 answers to your 32 buyer queries this week (Gemini excluded). Measured on your questions, not an industry average.
Set it up once. Deep scan every week. Read the trend.
Three steps, and only the first one takes any of your time. The setup guide walks that first step screen by screen, with real screenshots. The rest runs on a weekly cadence and delivers itself.
Choose what you get measured on
You pick the questions you want to be cited for, not generic SEO keywords. We draft the starting bank from your site, then you edit, add, reorder and pin the ones that matter. Run the free check on 8 sample prompts to see a snapshot first. SMB tracks 30 questions weekly; Agency tracks 40 per client domain, split across brand-defense and category questions.
The prompt bank is the measurement frame. Change it and you change what is measured, so it stays stable week to week and each new prompt is added alongside the old ones. That is what keeps the trend line comparable.
We run a real scan every week
We send each of your prompts to ChatGPT, Perplexity, Claude, Gemini, and Google's AI Mode, then read back the full ordered list of sources each answer cited. ChatGPT and Gemini answers are captured from each app as a logged-out user in your market, through a data provider, so you see what a real buyer sees. Claude and Perplexity answers come from their APIs, and Google's AI Mode from Google's own results. No model of what an engine might say.
We record whether you were cited, where you sat in the source list, which competitor held the slot when you did not, and how the answer framed you when you did. On Agency, every prompt runs 3 times per engine and the passes are averaged: 40 prompts x 5 engines x 3 passes = 600 real AI queries per client, every week.
Email lands. Trend builds.
The moment the scan finishes we email you a digest naming the engine and competitor that moved most this week. Your dashboard reads like Search Console for AI answers and stacks week over week, so the headline is the drift, not a one-shot verdict.
On the Agency tier the same weekly cycle produces a white-label client report on one durable link: verdict, movement, verbatim AI quotes, competing sources, and the plan. Numbers freeze per week, so a client can compare last week to this week honestly.
Identified. Fixed. Moved. The loop every lost answer runs.
The weekly scan names a question a rival holds. Fixes turns it into a short checklist. You tick the steps, confirm it is live, and the next scans show whether the engine flipped. This is one real fix from the sample client, with every number read off the same scans the report shows.
- Who wins it
- sprudge.com, an article, 7 weeks running
- Claude
- Not citedin 7 of 7 scans
- Your page
- /roast-fresh, answers it below the fold
The scan also keeps what Claude said and which page it quoted, so the fix starts from the answer the engine actually gave.
- Rewrite the first paragraph to answer the question directly
- State the roast date and the ship-same-week window as plain sentences
- Confirm it is live
Claude started citing /roast-fresh 2 weeks after it shipped, and has on 6 of the 6 weekly scans since.
Your confirmed week is marked on the Overview chart too.
Sample data, Fernway Coffee, a fictional brand. A flip is called only when the engine cites you on two consecutive weekly scans after you confirm. Work on your own page can land in a week or two; off-page work, a listing or a thread, usually takes 4 to 8 weeks, and the chart says so rather than guess.
Every engine cites differently. We measure all five.
A tool that only checks ChatGPT tells you a fifth of the story. Radar runs every prompt through all five mainstream engines, because the same brand can win in one and be invisible in another. Here is roughly what earns a citation in each.
One signal cuts across all five: brand mentions. Ahrefs' 2025 studies of 75,000 brands found brand mentions correlate with AI visibility at 0.66 to 0.71 across ChatGPT, AI Mode and AI Overviews, the single strongest correlate, well ahead of backlinks. Radar tracks that footprint too, so the report points you at the work that actually moves the number.
We measure citation the way the engines produce it. Ask the real question. Record the real answer.
Real query, real answer
We send your actual prompt to each engine and read back the full ordered list of sources it cited. ChatGPT and Gemini are read from their apps as a logged-out buyer in your market; the other engines through their APIs. No model of what it might say.
Honest when we cannot measure
If an engine returns nothing usable for a prompt, that shows as no data, never a fabricated zero. A steady week says steady rather than inventing movement. That is what makes the number defensible in front of a client.
Six signals, because citation is binary and the answer is not.
A yes-or-no cited rate is too shallow to drive a roadmap. Every weekly scan reports six distinct signals, each one a different lever.
1. Prompt intent coverage
Every tracked question carries a buyer stage, from problem-aware down to ready-to-buy. Being cited on trivia is not the same as being cited where buying decisions happen, so the rate is sliceable by stage.
In the app: the Stages cut in Slice the numbers on the Overview, and the stage on each query's deep-dive.
2. Recommendation rate
An answer can name you in passing or actually recommend you. Each appearance is classified from the stored answer text (top pick, listed option, mentioned, caveated, unfavorable), and on vendor-comparing and ready-to-buy questions we state the share where you were genuinely recommended.
In the app: the Appearance cut in Slice the numbers on the Overview.
3. Brand attribute framing
What the answers say you ARE: trusted, premium, technical, expensive, complex to set up. These labels are classified from the answer text and rolled up, because how AI search describes you is your positioning whether you chose it or not.
In the app: the engine cards on every query page, plus the rollup in the Appearance cut.
4. Source dependency
Which third-party pages the answers lean on: the Reddit thread, the review site, the competitor comparison page. Knowing who holds the slot tells you exactly where to earn presence instead of guessing.
In the app: Competitors & sources, down to the specific thread.
5. Stability
AI engines are probabilistic, so a result that flips week to week is noise to wait out, while a stably absent question is a real gap worth fixing. Each question carries a stability read derived from repeat runs, cross-engine agreement, and recent weeks. On Agency, every prompt also runs 3 times per engine and averages into a confidence band; no funded competitor in the category advertises a multi-run confidence measure.
In the app: the stability read under each query on the Overview leaderboard, and the confidence chip on the weekly verdict.
6. Gap mapping
A miss should point at a cause. Every uncited question names who holds the slot and where, and Fixes turns each lost answer into a checklist and each recurring pattern into a named action, including the specific objection the caveated answers keep raising.
In the app: the why-not reason on every uncited query, and Fixes.
A note on the bright line: this is measurement, not the scanner's prediction. The WhyIQ CRO Scanner's AI Citability Index scores a page for the on-page signals that make citation likely. AI Radar records the citations that actually happened. One predicts; the other measures.
Every week ends in something you can put in front of a client.
One durable white-label link per client. A plain-language verdict, what moved, what the AI actually said verbatim, who holds the answer slots you are missing, and the plan. Your branding, your name on it, numbers frozen per week. This is the real report component with sample data, not a mock.
What the AI actually said
Verbatim excerpts from this week's AI answers: where Fernway Coffee was cited or named, and where an answer sent the reader to a rival instead.
CitedThe AI answer linked to your site as a source. This is the win: the engine did not just know about you, it pointed the reader to you.ChatGPT answering "best specialty coffee subscription" “For freshness-focused buyers, Fernway Coffee is a strong pick: beans ship within 48 hours of roasting and the subscription lets you pause or reschedule from a single dashboard.”
CitedThe AI answer linked to your site as a source. This is the win: the engine did not just know about you, it pointed the reader to you.ChatGPT answering "Fernway Coffee reviews" “Fernway Coffee reviews consistently highlight the roast-date transparency and a subscription that is easy to pause, with most complaints limited to limited decaf options.”
A methodology footer states how the numbers were measured, with sample sizes. On Agency it reads: each question ran 3 times per engine and the results were averaged into a confidence band. A steady week says steady. That honesty is what makes the number defensible in front of a client.
How it works, straight answers.
How is this different from the scanner's AI Citability score?
They sit on opposite sides of the same line. The WhyIQ CRO Scanner's AI Citability Index predicts readiness: it audits a page for the on-page signals (FAQ content, statistical density, named author, crawler access) that make a citation likely. AI Radar measures the outcome: it runs your real prompts through the real engines and records the citations that actually happened. The scanner tells you if a page is ready to be cited. AI Radar tells you if it actually is, by which engine, and for which prompt.
Do you really run the prompts, or is this a prediction?
We run them. Every weekly scan sends each of your buyer prompts to ChatGPT, Perplexity, Claude, Gemini, and Google's AI Mode, and we read back the exact sources each answer cited. It is a measurement of what the engines actually said, not a model of what they might say. Zero rows when we genuinely cannot measure, never a hallucinated zero. ChatGPT and Gemini answers are captured from each app as a logged-out user in your market, the way a real buyer would see them; Claude and Perplexity come through their APIs.
Why weekly and not daily or real-time?
Two reasons. Off-page work (content, entity cleanup, review-site footprint) takes 4 to 8 weeks to land in AI answers, so the useful signal is the trend, not the minute-to-minute snapshot. And AirOps' 548,000-page study found AI citations have a roughly 3-month half-life, which makes weekly the minimum cadence that actually catches drift without drowning it in noise. A daily single-read tool reports noise as a verdict.
Why does the Agency tier run every check 3 times?
AI engines are probabilistic. Ask the same question twice and you can get two different answers, so a single read is a coin flip dressed up as a fact. The Agency tier queries every prompt on every engine 3 times each week and averages the passes into a confidence band, so the report reads 'cited 2 of 3 passes' instead of a yes-or-no that flips next week. If a number is going in front of a client, it should be one you can defend.
Which engines are included, and can I add more?
Five on every tier: ChatGPT, Perplexity, Claude, Gemini, and Google AI. There are no per-engine add-ons; the price is flat and every engine is in every plan. The Agency tier upgrade is rigor (the 3-pass confidence band) and client domains, not more engines. Grok and Microsoft Copilot are on the roadmap for the Agency tier once their measurement pricing stabilises.
See your snapshot in about three minutes. Then watch the trend build.
Run a free check on 8 sample prompts across all 5 engines, no account needed. Or look at a full sample client report first: fictional client, real product.