AI Search Visibility Audit: ChatGPT & Perplexity

Find out how often AI answer engines recommend your brand when buyers ask them for advice — and who they recommend instead. The Actor reads your website, writes 30 realistic buyer-intent questions, asks ChatGPT, Perplexity and/or Gemini each of them several times, and delivers a shareable HTML report with your brand's mention rate, citation rate, share of voice against competitors, the sources the engines rely on, and every sentence they wrote about you. This is an unofficial tool and is not affiliated with OpenAI, Perplexity, or Google.

What does the audit measure?

• Mention rate – share of AI answers that name your brand (or an alias or your domain), with a 95% confidence interval.
• Citation rate – share of answers whose sources include your website.
• Average position – where your brand appears among the brands named.
• Share of voice – your mentions versus each listed competitor.
• Most cited sources – the domains AI engines trust for your market (the places where you need to be present).
• What the engines say – every sentence about your brand, so you can catch outdated or wrong claims.
• By engine and by question type – best-of lists, comparisons, how-to, pricing, trust, and use-case questions are reported separately.

How it works

1. Crawl – the homepage plus up to five key pages (about, products, pricing, blog) are read to understand what you sell and to whom.
2. Prompt generation – an LLM (OpenAI if a key is given, otherwise Gemini) turns that profile into buyer-intent questions spread across six categories. Questions do not name your brand except in comparison / trust categories, so results reflect unprompted recall.
3. Sampling – every question is asked samplesPerPrompt times per engine (default 3), because AI answers vary from run to run. ChatGPT is queried through the OpenAI Responses API with web search when available, Perplexity through sonar, Gemini with Google Search grounding.
4. Analysis – mentions, citations, ranks, competitor mentions and cited domains are extracted from every answer; rates get Wilson 95% intervals.
5. Report – REPORT.html in the run's key-value store, plus one audit summary item and one promptResult item per answer in the dataset.

Example: what an audit surfaces

A September 2026 audit of apify.com (Gemini with Google Search grounding, 30 questions × 3 samples, competitors Bright Data, ScrapingBee, Zyte):

• Apify was named in 48% of answers (95% CI 38–58%), on average in position 1.4 when named; share of voice 36% vs Bright Data 33%.
• Only 18% of answers cited apify.com as a source. The domains the engine leaned on most were brightdata.com (29 answers), grepsr.com (26), octoparse.com (22) and scrapingbee.com (21) — competitor content, not the brand's own pages.
• use-apify.com, a third-party guide site that describes itself as "independent, not owned by Apify", was cited in 14 answers — almost as often as the official site (16). AI engines were sourcing Apify's pricing and product claims from a lookalike domain the brand does not control. That is exactly the kind of finding the Most cited sources table exists for: check what such sites say about you, and make sure the official pages give engines something better to cite.

How to use it

1. Enter your Website URL and Brand name; add Competitors to measure share of voice.
2. Paste at least one API key. Cheapest full audit: a Google AI Studio key with billing enabled — select the Gemini engine; a default 30 × 3 audit costs about $0.20 in Gemini 3.x Flash tokens and Google Search grounding stays within Google's 5,000 free searches per month (September 2026 prices). Questions are generated by Gemini too. A free (no-billing) key also works but is small — observed: 20 requests per day per model, no grounding — so set about 12 prompts × 1 sample there. Add an OpenAI key for the ChatGPT engine and a Perplexity key for Perplexity.
3. Keep 30 prompts × 3 samples for a first audit (about 90 answers per engine; about $0.20 on a paid Gemini key, typically under $1 of OpenAI usage with gpt-4o-mini). Rate limits are waited out automatically; when a provider quota runs out mid-audit the engine stops early and the report says so.
4. Click Start. Open REPORT.html from the Storage → Key-value store tab or via the link in the audit dataset item.

No API key? Enable Mock mode to see the full report format with simulated answers — mock runs are free and clearly labelled.

Input

Field	Default	Notes
websiteUrl	–	Brand homepage (required).
brandName	–	Brand as written in text (required).
brandAliases	–	Extra spellings or product names counted as mentions.
competitors	–	Up to 10 competitor names.
businessDescription	–	Optional context to improve prompt generation.
promptCount	30	5–60 generated questions.
samplesPerPrompt	3	1–5 answers per question per engine.
engines	["openai"]	openai, perplexity, gemini.
openaiApiKey / openaiModel	– / gpt-4o-mini	ChatGPT engine; generates the questions when present.
geminiApiKey / geminiModel	– / gemini-flash-latest	Gemini engine (Google Search grounding when the key has grounding quota); generates the questions when there is no OpenAI key. If the model is retired or has no quota, the newest available Flash model is used.
perplexityApiKey	–	Perplexity sonar engine.
mockMode	false	Simulated answers, free of charge.

Output

audit item (one per run):

{
  "type": "audit",
  "brand": "Apify",
  "engines": ["openai", "perplexity"],
  "promptCount": 30,
  "brandMentionRate": {"count": 49, "n": 180, "rate": 0.2722, "ci95": [0.2115, 0.3426]},
  "brandCitationRate": {"count": 31, "n": 180, "rate": 0.1722, "ci95": [0.1236, 0.2348]},
  "averageBrandRank": 1.6,
  "shareOfVoice": {"Apify": {"mentions": 49, "share": 0.31}, "Bright Data": {"mentions": 44, "share": 0.28}},
  "topCitedDomains": [{"domain": "g2.com", "count": 38}, {"domain": "apify.com", "count": 31}],
  "reportUrl": "https://api.apify.com/v2/key-value-stores/.../records/REPORT.html"
}

promptResult items (one per answer): engine, model, promptId, category, prompt, sampleIndex, answer, citations, brandMentioned, brandCited, brandRank, competitorsMentioned, citedDomains, brandSentences, error, latencyMs, mode.

Pricing

One audit event per completed live audit (the report and dataset are delivered first; a run that fails before the report is written is not charged, and a partial audit in which fewer than half of the planned answers succeeded — for example because a free-tier quota ran out — is delivered free). Mock runs are free. Your own OpenAI / Perplexity / Gemini API usage is billed by those providers separately: about $0.20 per default audit on a paid Gemini key, typically well under $1 on OpenAI, nothing on Gemini's free tier (small audits only).

Tips

• Run the audit monthly and compare brandMentionRate over time; the confidence interval tells you whether a change is real.
• Look at Most cited sources first: getting listed on those domains is the fastest way to raise your citation rate.
• Use businessDescription when your homepage is thin or JavaScript-only.
• Keep prompts at 30 and samples at 3 for comparable results across runs.

FAQ

Do I need to give you my API keys? Keys are entered as secret input fields, used only during the run, and never stored in the dataset or report.

Why do the numbers differ between runs? AI answers are non-deterministic; that is exactly why each prompt is sampled several times and rates carry confidence intervals. Larger samples narrow the interval.

Which models are used? OpenAI gpt-4o-mini by default (configurable), Perplexity sonar, Gemini gemini-flash-latest (configurable; the newest stable Flash) with Google Search grounding.

Can I run it without paying for an AI API? Yes, on a small scale: a Google AI Studio key has a free tier (observed September 2026: 20 requests per day per model, without Google Search grounding). Select only the Gemini engine, set about 12 prompts × 1 sample, and leave the other keys empty. A full 30 × 3 audit needs billing enabled on the Google project (about $0.20) or an OpenAI key.

Can it fix my visibility? It measures and points to the sources that matter; content and PR work is up to you.