AI Visibility API: brand mentions and citations in AI answers
The AI Visibility API asks AI engines the questions your buyers ask and returns, for every answer, whether your brand was named, which of your URLs were cited and at what rank, and which domains won where you are absent. One call covers up to 10 questions on ChatGPT search, Google AI Overviews and DeepSeek. On 26 October 2026 a two-question panel on all three came back in 12.7 seconds and cost $0.05.
$2 of free API usage per month · No card required · Rows with an error are never billed
$ curl https://api.quanticdata.io/v1/scraper/ai-visibility \ -d '{ "url": "https://quanticdata.io/serp-api/", "queries": ["best serp api for ai agents", "google search results api pay as you go"], "engines": ["openai", "aio", "deepseek"], "brand": "QuanticData", "country": "us", "no_render": true, "no_retrieval": true }' { "citations": { "rows": [ …, { "query": "google search results api pay as you go", "engine": "openai", "cited": true, "mentioned": true, "rank": 6, "ourUrls": ["…/collectors/google-search-results-api/…"] }, … ], "coverage": { "openai": { "asked": 2, "cited": 1, … } } }, "billing": { "citation_calls_billed": 4 }, "usage": { "cost_usd": 0.05 } } # 2 questions × 3 engines · 4 answers billed · 12.7 s
What is an AI visibility API?
An AI visibility API sends your questions to AI engines and returns structured data about the answers: who is named, which pages are cited, and how often you are among them. It is LLM visibility tracking you can script, instead of typing the same prompts into three chat windows and pasting the results into a spreadsheet.
QuanticData adds the other half of the question in the same call. Next to the citation panel, every request audits the page you name the way an AI crawler sees it: access for 24 AI user agents in your robots.txt, the text that exists without JavaScript, structured data, answer-shaped copy, dates and authorship. When you are not cited, the response already holds the reasons that can be fixed on your side.
What one call measures
Brand mentions
mentioned is true when the answer text names your brand, linked or not. Set brand to the name buyers use; it defaults to the site name.
Citations of your URLs
cited, ourUrls and rank: which of your pages the engine listed as a source, and where your domain sits among the distinct domains it cited.
Every source, in order
citedUrls and allDomains keep the engine's own order, so you see the page that won the answer, not only the site.
Share of voice
shareOfVoice counts every cited domain across the panel, with its share of answers, your domain flagged as ours and the domains you pass in competitors flagged too.
Where you lose
losingTo lists the domains cited in the answers where you are absent: the pages to study, or to get yourself mentioned on.
Per engine and country
coverage gives asked, cited and mentioned for each engine; country sets where the question is asked from, one market per call.
Which AI engines it checks
Three engines answer on the panel today. Each one is a different kind of answer, and the response says exactly what each is in citations.engineNotes, so nobody mistakes an API for the consumer app.
| engine | What answers | What you get |
|---|---|---|
openai | OpenAI's API with its web search tool, the closest programmatic equivalent of ChatGPT search (not chatgpt.com itself) | cited URLs in order, mention, rank |
aio | the Google AI Overview, read from a live Google results page in the country you choose | whether the overview names your brand, and its text; citations when the overview exposes its sources |
deepseek | Google's top 10 for the question, handed to DeepSeek with the instruction to answer and cite | which of the same results a model picks: cited URLs, mention, rank |
Pass engines explicitly. A row an engine could not answer, because Google showed no AI Overview for that question or because an engine is not available, comes back with an error field, is left out of coverage and share of voice, and is not billed.
How it works
From your questions to one row per answer
1. Send the page and the questions
A URL, up to 10 questions, the engines, your brand name and the competitor domains to watch. The URL's domain is what counts as yours in every answer.
2. Every question goes to every engine
The pairs run in parallel from the country you pick, while the page itself is audited the way an AI crawler fetches it.
3. One row per answer comes back
Mention, citations and rank for each question on each engine, then coverage, share of voice and losingTo across the panel. You pay for the answers obtained.
| Field | What it does |
|---|---|
url | required: the page to audit; its domain is matched in every answer |
queries | up to 10 questions; leave it out for the on-page audit alone |
engines | openai, aio, deepseek |
brand | the name to look for in the answer text, up to 80 characters |
competitors | up to 20 domains flagged in the share of voice |
country | two-letter code: where the question is asked from |
no_retrieval | skips the 2 searches that check where the page ranks for its own H1 question and whether it is indexed |
offsite | adds 5 searches for your brand on YouTube, Reddit, Wikipedia, LinkedIn and review sites |
Integration
One POST, one row per answer
curl https://api.quanticdata.io/v1/scraper/ai-visibility \
-H "Authorization: Bearer $QD_API_KEY" \
-H "Content-Type: application/json" \
-d '{ "url": "https://example.com/pricing/",
"queries": ["best invoicing app for freelancers",
"invoicing software with recurring billing"],
"engines": ["openai", "aio", "deepseek"],
"brand": "Example",
"competitors": ["rival-one.example", "rival-two.example"],
"country": "us", "no_retrieval": true }'
The field names in the Response tab are the real ones; the values belong to the example request. Every field, every error code and the full on-page audit are in the API reference.
The same question, asked twice
On 26 October 2026 we ran the API on our own SERP API page from the United States, with two questions. The run shown at the top of this page asked all three engines with no_retrieval set, so the bill is exactly the audit plus the answers.
| What we measured | Result |
|---|---|
| Answers billed | 4 of 6 rows: ChatGPT search and DeepSeek on both questions |
| AI Overview rows | Google showed no AI Overview for either question; both rows free |
| Cost | $0.05: $0.01 for the audit, 4 × $0.01 for the answers |
| Time for the whole call | 12.7 seconds |
| "google search results api pay as you go" on ChatGPT search | cited a quanticdata.io collectors page, 3rd of 6 domains in a first run and 6th of 6 in a second run minutes later |
The last row is the reason AI visibility tracking means repeating the panel. The same engine, asked the same question from the same country, cited the same page both times in a different position. One answer is a sample; a trend needs the same questions asked on a schedule.
Pricing
Per answer. No subscription.
Each call pays the audit once and then one price per answer obtained, where an answer is one question on one engine: the prices shown in the dashboard and the API reference. Before a panel runs, the API checks that your balance covers every pair; afterwards it bills only the rows that came back without an error.
Worked example: N questions × M engines
Cost of one run = $0.01 + N × M × $0.01, at most. Rows with an error lower it.
| Panel | Cost per run | Answers | Four runs a month |
|---|---|---|---|
| 1 question × 1 engine | $0.02 | 1 | $0.08 |
| 5 questions × 2 engines | $0.11 | 10 | $0.44 |
| 10 questions × 2 engines | $0.21 | 20 | $0.84 |
| 10 questions × 3 engines | $0.31 | 30 | $1.24 |
Ten questions on ChatGPT search and DeepSeek once a week, four runs a month, cost at most $0.84, which fits inside the $2 of free usage every account gets. By default each call also runs 2 searches to see where the page ranks for its own H1 question, billed as SERP API searches; no_retrieval: true gives exactly the figures above, and offsite: true adds 5. Monthly plans are on the pricing page.
Who uses an AI visibility API
SEO and GEO agencies
One call per client page and question set, the rows straight into the monthly report, and a cost per answer instead of a seat per client.
In-house content teams
Find the questions where you are absent and the pages that win them, then check after a rewrite whether the engines changed their minds.
Product builders
Put AI visibility data inside your own dashboard, CRM or alerting. The JSON is per answer, so you decide how to score it.
Give it to your agent as a tool
The QuanticData MCP server exposes this API as the ai_visibility tool, so Claude, Cursor or any MCP client can run a panel mid-conversation and reason over the rows. To try a request before writing code, open the AI Visibility Audit in the Playground. To see what a single page looks like to an AI crawler first, the free AI visibility audit runs the on-page pass with no account; the rest of the data layer for agents is in the web data API for AI.
QuanticData vs a typical AI visibility tool
| QuanticData AI Visibility API | Typical AI visibility tool | |
|---|---|---|
| How you pay | $0.01 per answer obtained, no subscription | a monthly plan sized by the number of tracked prompts |
| What comes back | JSON per question and engine: cited URLs in order, mention, rank | a dashboard; API access depends on the plan |
| Why you are not cited | the on-page audit runs in the same call | usually a separate audit |
| Free tier | $2 every month, renews forever | usually a time-limited trial |
| AI agent access | native MCP tool ai_visibility | a wrapper to write |
The setting to start with
POST /v1/scraper/ai-visibility with your page as url, the ten questions your buyers ask, engines: ["openai", "deepseek"], your brand, two or three competitor domains and your main country: 20 answers for $0.21 at most. Run it weekly from cron or n8n, and the $2 of free API usage per month covers it.
Sources and standards
What this product reads or implements, so you can verify a claim yourself:
What is an AI visibility API?
An AI visibility API sends your questions to AI engines and returns structured data about the answers: whether your brand is named, which of your URLs are cited and in what position, and every other domain the engine relied on. QuanticData does it in one POST for up to 10 questions on ChatGPT search, Google AI Overviews and DeepSeek.
Which AI engines does it check?
Three today: openai, OpenAI's API with its web search tool, the closest programmatic equivalent of ChatGPT search; aio, the Google AI Overview read from a live results page in your country; and deepseek, Google's top 10 for the question handed to DeepSeek to answer and cite. Rows an engine cannot answer are not billed.
What is the difference between a mention and a citation?
A mention means the answer text names your brand; a citation means the engine lists one of your URLs among its sources. The API reports both for every answer, because an engine can recommend you without linking you, or link one of your pages without naming the brand. The brand field sets the name to look for.
How much does the AI Visibility API cost?
$0.01 per audit plus $0.01 per answer obtained, where an answer is one question on one engine. Ten questions on three engines cost at most $0.31 a run. Rows that come back with an error, such as a question with no AI Overview, are not billed, and every account gets $2 of free usage a month.
Is there a free AI visibility API?
Every account with a permanent email address gets $2 of API usage free each month, with no card required, shared across all QuanticData Data APIs. That covers ten questions on two engines every week, with credit to spare. The free AI visibility audit tool checks a single page the way an AI crawler sees it, without an account.
Can I track share of voice against competitors?
Yes. shareOfVoice lists every domain cited across the panel with the number of answers that cited it and its share, and flags your domain and up to 20 competitor domains you pass in competitors. losingTo lists the domains cited in the answers where you are absent, which shows the pages that win the questions you are missing.
Can I check AI answers by country?
Yes, one country per call. country sets the proxy exit and the locale of the Google results page the AI Overview is read from, passes an approximate country location to ChatGPT search, and localizes the Google results DeepSeek answers from. To compare markets, run the same panel once per country and keep the responses side by side.
Is this the real ChatGPT?
It is OpenAI's own API with its web search tool, the closest a program can get to ChatGPT search; it is not chatgpt.com in a browser, and the response says so in engineNotes. Answers move between runs: the same question, asked twice on 26 October 2026, cited our page third and then sixth.
How often should I run it?
On the schedule your report needs: daily, weekly, or before and after you change a page. There is no scheduler inside the API, so call it from cron, n8n or an agent and store each response as a dated snapshot. Read a trend over several runs, because a single answer can change from one run to the next.
What are the limits?
Up to 10 questions per call, 20 competitor domains and a brand name of up to 80 characters. Each account can get 300 answers a day from ChatGPT search and DeepSeek together, counted as questions × those engines; AI Overview rows do not count. When the limit is reached the API answers 429 with the time it resets.
Can AI agents use it directly?
Yes. The QuanticData MCP server exposes the same call as the ai_visibility tool, so Claude, Cursor or any MCP client can run a panel mid-conversation and read the rows. From your own code it is one HTTPS POST with a Bearer key, and the Playground builds the request for you before you write any.
Run your first panel in a minute
$2 free every month, no card. Mentions, citations and share of voice from the first call.
Get my free API key