AI Brand Visibility Tracker — track your brand in ChatGPT, Gemini, Perplexity, Claude & Google AI Overviews

Your customers are asking AI for recommendations. Is your brand in the answers?

This actor checks how often AI answer engines mention and cite your brand — versus your competitors — for the buyer-intent questions that matter to you. It's the data layer for GEO / AEO (Generative Engine Optimization): the same insight dedicated SaaS visibility tools sell for $99–500/month, priced per check instead.

What you get

For every prompt × platform combination:

• ✅ Was your brand mentioned? At what position relative to other brands?
• 🔗 Which sources did the AI cite — and was your website among them?
• ⚔️ Which competitors were mentioned, and where?
• 📊 Aggregates: visibility score (0–100), share of voice, citation rate, top cited domains
• 📈 Persistent trends: automatic comparison with the previous run and a 12-run history
• 🎯 Priority gaps: the prompts competitors win and the sources cited in those losing answers
• 🎲 Optional 2–3 sample confidence checks so one variable AI answer is not mistaken for a fact
• 📄 A clean HTML report you can forward straight to a client or your team

Supported AI platforms

Bring-your-own-key platforms run on your quota — your key is used only for your checks and never stored.

Quick start

1. Enter your brand name and domain (e.g. apify.com).
2. Add 5–20 prompts — the questions your customers actually ask AI ("best CRM for small agencies", never your own brand name). No prompts yet? Leave the field empty and fill in Category instead ("CRM for small agencies") — the actor generates a starter set for you.
3. List your competitors (optionally with domain: Zyte|zyte.com). These sharpen generated prompts too: competitors become "alternatives to Zyte" and "Zyte vs Octoparse" — the questions where an AI can name you without being asked about you.
4. Optionally set a Country (two-letter code like de) and Language to measure a specific market instead of the US/global default.
5. For less noisy measurements, choose 2–3 Samples per prompt (each sample counts as a check). One is the economical default.
6. Keep trend history enabled. Reuse the same optional Monitor ID on scheduled runs; when blank, the actor derives one automatically.
7. Pick platforms and hit Start. When the run finishes the HTML report opens first in the Output tab — the dataset, SUMMARY and OPPORTUNITIES sit behind the same dropdown.

Output

The report opens first. When the run finishes, the Output tab shows the readable HTML report — visibility score, per-platform mention and citation rates, share of voice and the prioritized competitor gaps — with no clicking around. Every dataset row also carries a reportUrl in the last column, so an exported CSV still leads back to the report.

That link opens without an Apify login. It is a signed URL with no expiry, which is what makes it work in a spreadsheet — and it means sharing an exported dataset shares the report with it. Treat the file the way you would treat the report itself.

Each check becomes one dataset item:

{
  "prompt": "best web scraping platform for developers",
  "platform": "gemini",
  "model": "gemini-3.6-flash",
  "sample": 1,
  "samplesPerPrompt": 2,
  "checkedAt": "2026-07-11T12:30:00.000Z",
  "location": { "countryCode": "de", "language": "de" },
  "latencyMs": 4210,
  "brand": {
    "name": "Apify",
    "mentioned": true,
    "position": 1,
    "mentionCount": 3,
    "firstMentionSnippet": "…platforms like Apify offer…",
    "domainCited": true
  },
  "competitors": [
    { "name": "Zyte", "mentioned": true, "position": 2, "mentionCount": 1, "domainCited": false }
  ],
  "totalBrandsDetected": 4,
  "citedUrls": ["https://www.reddit.com/r/webscraping/…", "https://apify.com/…"],
  "citedDomains": ["reddit.com", "apify.com"],
  "citedSources": [{ "url": "https://www.reddit.com/r/webscraping/…", "domain": "reddit.com" }],
  "answerText": "…full answer for auditing…",
  "error": null
}

location appears when you set a country/language; failed checks keep the same shape with error: { "message", "code" } filled in, and Google AI Overviews items add aiOverviewPresent. The exact model, checkedAt, sample and latencyMs on every item keep historical comparisons meaningful even as providers update their models. The run key-value store holds REPORT (self-contained HTML, and the Output tab's default view), SUMMARY (JSON aggregates), OPPORTUNITIES (prioritized competitor/source gaps) and HISTORY (the monitor's rolling snapshots, written only when trackTrends is on). Persistent history is also kept in the named ai-visibility-history store so it survives across runs.

What a run costs

You pay a small start fee plus a fixed price per completed check (one prompt on one platform for one sample) — failed checks are never charged. So 10 prompts × 2 platforms × 1 sample = 20 checks; 3 samples would be 60 checks. The exact per-event prices are listed in the pricing section of this page, and every run logs its estimated charge up front before the checks start.

In practice that means a typical monitoring setup — 20 prompts across 2 platforms, checked weekly — costs about $9/month. Track all 5 platforms weekly and it's still around $22/month, versus the $99–500/month that dedicated AI visibility SaaS tools charge for the same data. You can cap total spend with Apify's built-in Maximum cost per run setting. Bring-your-own-key platforms additionally use your own provider API quota (billed by that provider, not by us).

Track trends with scheduled runs

AI answers change constantly — a single run is a snapshot. With trend history enabled, each run automatically compares its score, mention rate, citation rate, and prompt-level gains/losses with the previous run. The actor retains the newest 12 snapshots per monitor. Create a Schedule in Apify Console (weekly works well), point it at this actor with the same input/Monitor ID, and the trend section will populate automatically.

FAQ

Is this the same as asking the ChatGPT / Gemini / Claude app directly? Close, but not identical — and we'd rather you know exactly what's measured. Checks run through each platform's official API with web search or grounding enabled (the same underlying answer engines), not through a logged-in consumer app. Consumer apps add personalization, account history, and regional experiments that no monitoring tool can fully reproduce. Treat the results as a strong proxy for what a typical signed-out user is told. For full auditability, every check records the exact model that answered (model field in the dataset, also shown in the report) plus the complete answer text.

Why do results differ between runs? AI answers are non-deterministic and personalized: the same prompt can produce different answers minutes apart, and Google doesn't show an AI Overview for every query every time. That variance is real data — track trends across scheduled runs rather than reading a single snapshot as ground truth.

How many samples should I use? Use 1 for cheap discovery and routine weekly monitoring. Use 2 or 3 when establishing a baseline, validating an important change, or presenting results to a client. The report shows sample agreement and flags prompt/platform combinations whose mention result changed between samples. Samples multiply both checks and run cost.

Can I check visibility in a specific country or language? Yes. Set Country to a two-letter code (e.g. de, gb, br): Google AI Overviews then searches Google from that country, and ChatGPT, Perplexity and Claude localize their web search to it. Set Language to control the language of Google AI Overviews results. For the AI chat platforms the answer language follows your prompts, so write prompts in your target language. One honest caveat: the Gemini API doesn't support location targeting yet, so Gemini checks stay global. Leave both fields empty and everything behaves exactly as before (US/global).

What prompts should I use? Questions with buying intent where you want to be recommended: "best X for Y", "X vs Y", "how to solve [problem your product solves]". Avoid prompts containing your brand name — you're measuring unaided visibility.

Can I run it without writing prompts? Yes. Leave Prompts empty and set a Category (e.g. project management software). The actor builds a starter set from your category and competitors, prints it in the log before spending anything, and measures that. Two properties are deliberate. The set is deterministic — the same input always yields the same prompts, so scheduled runs stay comparable instead of resetting your trend line every week. And it never contains your brand name, because a prompt that names your brand guarantees a mention, and mention rate is 60% of your visibility score — a generator that asks "alternatives to [your brand]" is grading its own homework. Treat the generated set as a starting point: once you see which questions matter, paste them into Prompts and edit from there.

How is the visibility score calculated? Transparently: visibilityScore = round(60 × mention rate + 25 × citation rate + 15 × position score), where mention rate is the share of successful checks that mention your brand, citation rate is the share where your domain is among the cited sources, and position score averages 1/position over the checks where you're mentioned (1st brand mentioned = 1.0, 2nd = 0.5, …). Mentions weigh most because being in the answer is the battle; citations and early positioning refine it. The formula is versioned (scoreVersion in SUMMARY) — if the weighting ever changes, the version number changes with it, so trend lines stay honest.

What does "mentioned" vs "cited" mean? Mentioned = your brand name appears in the answer text. Cited = your website is one of the sources the AI links to. They're different levers: mentions come from being part of the conversation (reviews, comparisons, communities); citations come from having content AI engines consider a source of truth.

Is my API key safe? Key fields are marked secret in the input schema: Apify encrypts them at rest and they are decrypted only inside your actor run. The actor uses your key solely to call the provider you selected, during your run, and never writes it to logs, datasets, or reports. (To be precise: the actor code does receive the plaintext key at runtime — that's how any BYOK integration works — it just never persists it anywhere.)