AI Search Visibility Action Plan

Can ChatGPT and Perplexity find your business? Run one report and see what they say about you, what is stopping them, and the three things to fix first.

Give it your domain and your brand. It checks whether the AI search crawlers can fetch and read your site, asks ChatGPT and Perplexity your buyers' questions on your own API keys, and hands back a ranked fix list where every recommendation points at the evidence behind it. One domain per run, pay per report, no subscription.

Who it's for

• Marketing and SEO leads who are asked "why aren't we in ChatGPT's answer?" and need something better than a guess.
• Agencies producing an AI-visibility audit for a client — the report is the deliverable, with the evidence attached.
• Founders and product marketers checking a launch site before the first campaign.

It is for domains you control or are allowed to audit.

What you get

Three tiers, in the order to act on them:

1. Can they reach you? — the technical checks (what the Actor calls eligibility): which AI crawlers your robots.txt allows (search bots told apart from training bots — blocking GPTBot is a policy choice; blocking OAI-SearchBot hides you from answers), whether a WAF challenges the search bots while browsers pass, whether the main content is in the HTML or needs JavaScript (the crawlers don't run it), noindex/nosnippet directives, canonicals and redirect chains, Organization JSON-LD, sameAs profile links, the sitemap, and llms.txt (reported, not scored).
2. What they said about you — what the Actor calls observed inclusion: for each prompt and each engine you gave a key for: was the brand mentioned, was your domain cited, which competitors were named, and which sites the engine cited instead. Rolled up into mention rate, citation rate, share of voice and the top cited hosts per engine. An engine you did not query is absent, never zero.
3. What to do first — what the Actor calls inferred opportunity: where the engines go for answers on your topics and you are not yet: directories and review sites to get listed on, prompts where a competitor is named and you are not.

Outputs: OUTPUT (the report as JSON), REPORT.md (the action plan for people), and a dataset with one row per finding, observation and recommendation. Every recommendation carries the ids of the findings it rests on; the Actor refuses to write a plan that doesn't.

How the check works

The technical layer fetches your robots.txt, your sitemap, the homepage and up to five more pages (or the pages you name, up to ten) — once as a normal browser and once as each AI search bot (OAI-SearchBot, PerplexityBot, Claude-SearchBot) — and compares what each received: a 403, a challenge page or a body a fifth the size is a block. It then reads the raw HTML the way a crawler does, without running scripts: how many words are actually there, the title and description, robots directives in the tag and the header, the canonical, the JSON-LD, the profile links. At most 60 requests, two per second, no headless browser.

The engine layer sends each prompt to Perplexity's Agent API (the tier's preset, with web search) and to OpenAI's Responses API with its web-search tool, on your keys, and keeps the answer text, the citations and the token usage. It looks for your brand and your competitors in each answer and for your domain among the citations. The plan is a fixed rule table over those findings and metrics — no model writes the recommendations, so the same site with the same answers gets the same plan.

Don't know your topics? Leave them empty. The Actor reads your homepage and asks your own engine key for the things a customer would search for, keeps four, and where the homepage gives too little it fills the rest with questions about your brand by name — the report says which topics came from where. Tick Just suggest topics to see them first without buying a report.

Which AI should we ask?

One report asks one tier of models — the same tier on both engines:

Tier	Perplexity	OpenAI	When
Quick	fast	gpt-5.6-luna	The cheapest models on your keys. A first look.
Standard (default)	medium	gpt-5.6-terra	The models closest to what the ChatGPT and Perplexity apps show most people.
Deep	high	gpt-6-astra	Their strongest models — what a paying app user may see.

Answers come from the engines' official APIs — the same model families the apps use — without a person's history or settings, so what one customer sees can differ. To see what another tier says, run another report; the report price is the same, and the bigger models cost more on your own keys.

Input

Field	Meaning
domain	The site to check, as a hostname or URL. One per run.
brand	The name the engines should mention, as customers write it.
aliases	Other spellings or short names that count as a mention (up to 5).
competitors	Names to look for in answers, optionally `name
topics	What you want to be found for (up to 10); each becomes five prompts. Leave empty and the Actor reads your homepage and chooses four — an engine key is needed to ask them.
suggestOnly	Reads your homepage, writes the topics it would ask about, and stops. No report is charged; a key you give makes one small call. Paste the ones you want into topics and run again.
locations	Places added to the questions made from topics you typed, e.g. 'Taipei' (up to 3). Not used when topics are chosen for you.
prompts	Your own questions for the engines. 20 are included in the report price; each prompt above 20 is an extraPrompts event; 100 at most.
pages	Pages on the domain to check besides the homepage (up to 10). Empty → the first five sitemap URLs.
modelTier	Which tier of models to ask — quick, standard (default) or deep. See above.
perplexityApiKey	Queries Perplexity with your key. Stored as a secret, never written to the output.
perplexityModel	API only, not on the form: a specific Perplexity preset or model id that overrides the tier for this engine. Empty means the tier decides.
openaiApiKey	Queries OpenAI with your key. Stored as a secret, never written to the output.
openaiModel	API only, not on the form: a specific OpenAI model id that overrides the tier for this engine. Empty means the tier decides.
includeEvidence	Include short page and answer excerpts in the output (default on).
managed	Operator-only; refused on the public Actor.

Without any engine key the report covers the eligibility tier only, and is still one report.

Pricing

Pay per report, charged only when the report is written:

Event	Price	What it covers
byokReport	US$4.95	One domain, one model tier of your choice, 20 prompts on your own engine keys
extraPrompts	US$0.10	Each prompt above the 20 included, up to 100
managedReport	US$14.95	The same report on our engine keys — not yet available on the Store

Your engine keys are billed by Perplexity and OpenAI directly, at their list prices, for the prompts the report asks; the report's costs block shows the tokens and searches each engine used. A run that fails — an invalid domain, a rejected key — charges nothing.

Limits — please read

• One domain per run; up to 10 extra pages; up to 100 prompts.
• At most 60 requests to the site per run, two per second, ten seconds each; no headless browser — a page that needs JavaScript to show its content is reported as such, not rendered.
• Two engines in this version: Perplexity and OpenAI. Google's AI Overviews have no API and are not covered; the report says so rather than inventing a number.
• Answers vary between runs: the engines are not deterministic. The technical checks are — re-running them on the same site gives the same verdicts.
• The report is for domains you control or are allowed to audit.
• Chosen topics come from your homepage, four at most; where it gives too little, questions about your brand by name fill the rest — the report says which is which.

Compared with alternatives

	This Actor	AI-visibility SaaS (monthly)	A generic SEO crawl
Price	Per report, from US$4.95	US$30–500 a month	Per crawl or monthly
Crawler eligibility with evidence	Yes, per bot, search bots told from training bots	Rarely	No
What the engines actually answered	Yes, on your keys, per prompt	Yes, on theirs	No
Ranked plan tied to findings	Yes, by rule; every fix cites its evidence	Varies	Generic issues list
Runs from code or a schedule	Yes (Apify API)	Sometimes	Sometimes

FAQ

Do you need my API keys? Only to ask the engines. They are isSecret inputs — encrypted at rest, never written to the dataset, the report, the log or an error message. Leave them out and you get the eligibility tier only.

Why is Google not here? Google's AI Overviews and AI Mode have no API. Gemini's grounding API is a different product. Rather than pass one off as the other, the report covers what can be observed.

Does llms.txt help? Nobody has shown that the answer engines read it. The report records whether you have one and scores nothing on it.

Is this an SEO audit? No. It checks the things that decide whether an AI engine can fetch, read and cite you, then what the engines actually said, then where to earn a presence. A crawl tool lists issues; this ranks fixes by impact and effort with the evidence attached.

How current are the answers? As of the run. Re-run monthly, or after you ship a fix, and compare the metrics.

Support

Questions, a site the checks misread, a shape change at an engine: open an issue on this Actor's Apify page or write to hello@buildonmarketplaces.com. We answer within two working days. Built in public by BuildOn Marketplaces; the page for this Actor, with a real report on our own domain, is at https://buildonmarketplaces.com/solutions/ai-search-visibility.

Changelog

• 0.1 — First version: eligibility checks (robots per bot, crawler fetch, content, metadata, structured data, entity signals, sitemap, llms.txt), Perplexity and OpenAI on your keys, the ranked plan, OUTPUT/REPORT.md/rows.