Guides
Who uses AI visibility APIs: agencies, brands and SaaS platforms
Six kinds of team buy AI answer data, and the unit is the same: 37 of the 55 APIs we rate publish a rate per 1,000 answers, from $0.75 to $206.
By Mathias C.SEO consultant, and more recently GEO, at MentionBox
Six kinds of team buy AI visibility data through an API, and they are billed in the same unit: the AI answer collected. Agencies report on a book of clients, brand teams at large companies cover every market they sell in, SaaS platforms put GEO figures inside a product their own customers pay for, in-house teams at one brand track a short prompt list, e-commerce teams read the product cards assistants show, and developers feed figures into tools that already exist. Of the 55 AI visibility APIs this site rates, 37 publish a rate per 1,000 answers, from $0.75 with SERPHouse to $206 with Gumshoe, checked on October 23, 2026, so the first question for any of the six is how many answers a month its job consumes.
This page is for whoever has to work out which of those groups they belong to before shortlisting a vendor, and it covers the job each group does with the data, the monthly answer count behind it, the published clause or limit that settles the choice, the two groups the vendor pages tend to leave out, and what collecting the answers yourself on the engines' own APIs costs. Figures for the APIs come from each vendor's pricing page and documentation as recorded in our comparison, checked on October 23, 2026; the answer counts are our arithmetic on the schedules named in each section.
The six groups side by side
| Who | What they need answered | Monthly answers at a realistic schedule | What they buy | The published limit that decides |
|---|---|---|---|---|
| Agency reporting for several clients | where each client stands against its competitors, month after month | 3,200 for 8 clients × 25 prompts × 4 engines, weekly | computed figures plus the right to charge for them | 9 of 55 APIs publish a resale or white label right |
| Brand team at a large company | which engines and markets name the brand, with which sources | 11,520 for 120 prompts × 4 markets × 6 engines, weekly | wide engine coverage and a warehouse export | 5 of 55 cover all ten engines we track |
| SaaS platform embedding GEO data | what the feature costs per account, and whether it can be sold | 180 per account for 15 prompts × 3 engines, weekly | a published rate per answer and unattended delivery | rate limits per key, from 600 an hour to 300 a minute |
| In-house team at a single brand | whether the brand's own pages and name appear at all | 480 for 30 prompts × 4 engines, weekly | the free first-party reports, then a small usage-priced API | Search Console publishes impressions with no clicks |
| E-commerce merchandising team | which products assistants recommend in the category | 1,200 for 50 product prompts × 6 engines, weekly | answers carrying shopping product cards | 22 of 55 return shopping products |
| Developer or data team | the same figures inside tools the company already runs | varies with the job it supports | a self-serve key, SDKs, webhooks, an MCP server | 48 of 55 publish an MCP server, 15 publish webhooks |
API counts from the 55 AI visibility APIs compared on this site, checked on October 23, 2026. Monthly answer counts are our arithmetic on the schedule in the same row: prompts × engines × markets × runs per month.
Two columns in that table do most of the work. The answer count is what every usage-priced vendor bills, so it converts a brief into a budget before any demo. The published limit removes vendors from a shortlist for reasons no feature list shows: a resale ban, a missing engine, an hourly cap on a product that runs unattended. The e-commerce row is the narrowest of the six, since only 22 of the 55 APIs return the product cards an assistant shows beside its text, which our ranking for e-commerce brands scores on returned data at 40%.
Agencies: the terms settle it before the payload does
An agency collects one series per client and presents it under its own name, so the figures land in the report it already sends rather than in ten vendor dashboards each client has to log into.
The deciding clause is resale. Of the 55 APIs we rate, 9 publish a resale or white label right, 34 forbid resale or sublicensing, and 12 say nothing in either direction, checked on October 23, 2026. Three of the nine also return share of voice per brand, which is the figure a client report is usually built on: API for GEO, AthenaHQ and LLM Pulse. Our ranking of AI visibility APIs for agencies keeps only the APIs whose own pages allow it, and our guide on tracking ten clients without seat fees goes through the per-client arithmetic.
Eight clients on 25 prompts each, four engines, one country, refreshed weekly, is 3,200 AI answers a month:
| Route for 3,200 answers a month | Monthly usage | What comes back |
|---|---|---|
| DataForSEO at $1.20 per 1,000 | $3.84, after a $50 minimum payment | answer text and cited URLs, nothing computed |
| cloro at $4 per 1,000 | $12.80, inside a $30 Lite plan | answer text and cited URLs from real browser sessions |
| API for GEO at $13 per 1,000 | $41.60, pay as you go | share of voice, position, sentiment and ranked sources per brand |
| OtterlyAI at $15.75 per 1,000 | $50.40 of usage, inside a $189 Standard plan | computed metrics, capped at 2,000 API requests a month |
| LLM Pulse at €30.67 per 1,000 | €98 of usage, inside a €299 Scale plan | computed metrics on eighteen engines |
Rates from each vendor's pricing page, checked on October 23, 2026. Usage is our arithmetic: 8 × 25 × 4 × 4 = 3,200 answers.
At that volume the plan floor usually costs more than the usage: a book of eight clients consumes $41.60 of answers at $13 per 1,000, and the same book sits inside a $189 or €299 subscription elsewhere. The cheap route carries its own bill, since the scraping APIs return the answer text and leave share of voice, position and sentiment to code the agency writes and maintains. Our guide on share of voice, visibility and position sets out what those fields have to contain.
Brand teams: coverage first, then the warehouse
A brand team at a large company has a different constraint. It reports in every market where the brand sells, next to figures the vendor does not hold, which pushes the data into a warehouse and the choice toward whichever API covers the most engines. Five of the 55 APIs we rate cover all ten engines we track, and an engine a vendor skips shows up as a blank column in every market report for the rest of the year.
The volume follows the market count. One hundred and twenty prompts in 4 markets on 6 engines, refreshed weekly, is 11,520 answers a month, which is $149.76 at $13 per 1,000, $181.44 at $15.75 and $943.95 at AthenaHQ's $81.94. Our guide for enterprise brand teams prices a wider schedule across twelve APIs, and our enterprise ranking weights coverage and returned data at 35% each.
Procurement adds two questions the smaller buyers never ask. The first is where the API gets its answers, since a figure presented to a board has to survive the question of what was sampled: 24 of the 55 APIs collect from the live consumer interface, and 16 do not publish the method at all. The second is whether a nightly export can run inside the published limits. Profound's REST introduction states the API is available on the Enterprise plan, sets 600 requests an hour per key, returns X-RateLimit-Limit, X-RateLimit-Remaining and X-RateLimit-Reset headers, and offers /stream variants of its v2 visibility, citations, sentiment and query-fanout reports over Server-Sent Events. Scrunch's API reference states instead that its REST API "gives you the same data and configuration as the Scrunch dashboard", on a key an organisation admin creates.
SaaS platforms: a rate per answer and an API that runs alone
A SaaS platform that adds AI visibility to its own product is buying wholesale. The feature has to cost less per account than the plan it sits in, and nobody is watching the collection at three in the morning.
Three published facts decide the shortlist, in this order:
- A rate per answer, published. Thirty-seven of the 55 APIs we rate publish one; the other 18 quote on request, which means the margin on a $20 add-on cannot be worked out before a sales call. Fifteen prompts on three engines refreshed weekly is 180 answers per account per month: $0.36 at Scrapingdog's $2 per 1,000 for raw answers, $2.34 at $13, $14.75 at $81.94. Three hundred accounts on that schedule is 54,000 answers a month.
- The right to charge for it. Embedding metrics in a product you bill for is the rarest of the white label permissions, and our guide on white label AI visibility APIs reads the nine published clauses one by one. Our ranking for SaaS platforms keeps the APIs that allow resale and publish a usage price.
- Limits sized for a machine. LLM Pulse's REST overview publishes 300 requests a minute per key, up to 60 of them writes, with MCP calls sharing that budget and "no monthly or daily request limit", plus official SDKs in twelve languages. Profound's 600 requests an hour is a different shape of integration: a nightly export has to be paced across the hour rather than fired in one burst.
Fifteen of the 55 APIs publish webhooks, which is what turns a product feature into something that updates itself rather than polling. The rest leave you a scheduled job and a poll loop against the rate limit above.
The two groups the vendor pages leave out
A single brand with an in-house team
A brand tracking one market and one competitor set is the buyer most likely to be sold something it does not need yet, because part of the answer is published for free. Google's generative AI performance report has covered all websites worldwide since August 31, 2026 and shows impressions in AI Overviews and AI Mode, with no clicks, no queries and no position, under the same 1,000-row limit as the search performance report. Bing Webmaster Tools added intents, topics, citation share and period comparison on June 16, 2026, with citation share defined as "the percentage of citations attributed to your site" for a grounding query, and states plainly that it "does not expose competitor domains, represent traffic share, or assign quality scores to content".
Those reports answer whether Google and Microsoft surfaces cite your pages. They do not say whether your brand name was spoken in an answer, which competitor was named in your place, or what happens on ChatGPT, Perplexity and Claude. A paid API starts earning its keep at the point the competitor column appears in the brief. Thirty prompts on four engines refreshed weekly is 480 answers a month, which is $6.24 at $13 per 1,000, so the bill for this group is usually a plan minimum rather than usage. Our ranking of the cheapest APIs per 1,000 answers covers that end of the market, and our explainer on what AI visibility measures and what it misses covers what the percentage can carry.
Developers and data teams, who consume rather than report
The sixth group buys nothing on its own account. It serves one of the other five, and judges an API on tooling: a key from the signup page, a readable rate, SDKs, webhooks, an OpenAPI file, an MCP server. Forty-eight of the 55 APIs we rate publish an MCP server, the most widely shipped piece of tooling in this market, and our ranking for developers weights price and integration at 35% each. The job itself is plumbing: one scheduled collection, one table, several readers, which our guide on building an in-house GEO dashboard walks through.
Collecting the answers yourself, and who that suits
A team with engineers often asks why it should not call the engines directly. The search tools are published and priced, so the arithmetic is public. For 10,000 answers a month:
| Route | Published rate | 10,000 answers a month |
|---|---|---|
| OpenAI web search tool | "$10.00 / 1k calls + Search content tokens billed at model rates" | $100, plus tokens |
| Anthropic web search tool | "$10 USD per 1,000 searches", plus token costs | $100, plus tokens |
| Grounding with Google Search, Gemini 3 and newer | $14 per 1,000 requests after 5,000 free a month, capped at 1,500 requests a day | $70, within the daily cap |
| Grounding with Google Search, Gemini 2.5 | "$35 / 1,000 grounded prompts" | $350 |
| DataForSEO AI Optimization | from "$0.0012 per result page" on the LLM Scraper, "$0.0002 per task + LLM cost" on LLM Responses | $12 of scraper tasks |
Rates read on the official pricing pages on November 21, 2026; DataForSEO's AI Optimization rates also appear in our comparison at $1.20 per 1,000 answers, checked on October 23, 2026.
The limit sits in what those endpoints return. A grounded model API answer comes from the model you requested with the parameters you set, which is a different object from the answer the ChatGPT app, Google AI Overviews or Copilot gives a logged-out user, as our comparison of the OpenAI web search tool and ChatGPT answers sets out. The engine also decides whether to search at all: Anthropic's documentation states that Claude "answers directly without searching when the request draws on stable knowledge" and that triggering "is steerable through your system prompt", so an answer written from model memory and an answer assembled from fresh search results arrive through the same endpoint.
None of the three returns a visibility figure either. Brand matching, aliases, position inside the answer, share of voice and sentiment are code somebody writes and then maintains against a prompt list, a brand list and ten engines. Collecting it yourself suits a team that wants raw answers for an analysis no vendor models.
Mistakes we would avoid
- Buying seats for people who only read. The readers of a report are the cheapest part of this market. The answer count is the bill.
- Quoting a client from a prompt count. Thirty prompts on four engines in two markets at a weekly refresh is 960 answers a month, not 30.
- Signing before reading the resale clause. Thirty-four of the 55 APIs we rate forbid resale or sublicensing. A client-facing deliverable built on one of those is a contract problem, not a feature gap.
- Pointing a product at an hourly rate limit. Six hundred requests an hour per key is a comfortable analyst allowance and a tight ceiling for an unattended integration serving hundreds of accounts.
- Paying for answers you then recount. A scraping API at $1.20 per 1,000 is ten times cheaper than a computed one and leaves share of voice, aliases and sentiment to your code. That trade is reasonable once, and expensive to maintain across ten engines.
- Treating one run as a measurement. cloro's published study of August 14, 2026 puts about 300 answers per period at roughly ±5 percentage points, which is the floor any of the six groups should hold a report to.
Our view
In our view the six groups reduce to two questions. Who sees the number decides the contract: anything that reaches a paying client or an end customer needs the resale clause read first, which cuts 55 APIs to 9. How wide the brief runs decides the bill: prompts × engines × markets × runs, priced per answer, a figure you can produce in a minute and which most briefs in this market get wrong by a factor of the engine count.
API for GEO is a readable example of an API whose published facts answer both questions at once: resale and white label as an intended use, $13 per 1,000 answers down to $7 at its largest tier, pay as you go with no published minimum, share of voice, position, sentiment, ranked sources, themes and per-brand perception in one response, and reads over stored answers not billed. MentionBox, which publishes this site, has a commercial link with API for GEO. Its price per answer is roughly ten times a scraping API's, which is the recurring trade here: pay once for computed figures, or pay less and own the counting code.
Frequently asked questions
Who uses AI visibility APIs?
Six groups, with six different reasons. Agencies collect answers for several clients and put the figures in reports they charge for. Brand teams at large companies cover every engine and market they sell in and push the numbers into a warehouse. SaaS platforms embed GEO metrics inside a product their own customers pay for. In-house teams at a single brand track a short prompt list beside their SEO reporting. E-commerce teams read the product cards assistants show, which 22 of the 55 APIs this site rates return. Developers and data teams wire the figures into tools that already exist, which is why 48 of those 55 publish an MCP server.
What do agencies use an AI visibility API for?
To collect one series per client and present it under their own name, without buying each client a login on a vendor platform. Eight clients tracked on 25 prompts each, across four engines, refreshed weekly, is 3,200 AI answers a month, which costs $41.60 at $13 per 1,000 or $3.84 on DataForSEO's scraping queue at $1.20 per 1,000 with nothing computed. The clause to read first is resale: 9 of the 55 APIs we rate publish a resale or white label right, 34 forbid it and 12 are silent, checked on October 23, 2026.
Why would a SaaS platform buy an AI visibility API instead of building one?
Because the feature has to cost less per account than the plan it sits in, and a published rate per answer is what makes that arithmetic possible before signing. Fifteen prompts on three engines refreshed weekly is 180 answers per account per month, which is $2.34 at $13 per 1,000 and $14.75 at AthenaHQ's $81.94. The second constraint is the rate limit: LLM Pulse's REST overview publishes 300 requests a minute per key with no monthly or daily cap, while Profound's documentation sets 600 requests an hour per key, which an unattended product can exhaust in minutes.
Does an enterprise brand team need an API, or is a dashboard enough?
The dashboard stops being enough when the figures have to appear next to data the vendor does not hold, in every market the brand sells in. A brand tracking 120 prompts in 4 markets on 6 engines at a weekly refresh buys 11,520 AI answers a month, which is $149.76 at $13 per 1,000 and $943.95 at $81.94. Coverage is the other reason: 5 of the 55 APIs this site rates cover all ten engines we track, so an engine a vendor skips is a blank column in every market report.
Can you get AI visibility data for free?
For your own pages, partly. Search Console's generative AI performance report, live for all sites worldwide since August 31, 2026, shows impressions in AI Overviews and AI Mode with no clicks, no queries and no position, under the same 1,000-row limit as the search performance report. Bing Webmaster Tools added intents, topics, citation share and period comparison on June 16, 2026, and states that citation share "does not expose competitor domains, represent traffic share, or assign quality scores to content". Neither report says whether your brand was named in an answer, or which competitor was named instead.
What does one AI answer cost to collect yourself on the model APIs?
Ten thousand answers a month costs $100 of tool calls on OpenAI, whose pricing page lists web search at $10.00 per 1,000 calls plus search content tokens billed at model rates, $100 on Anthropic at $10 per 1,000 searches plus token costs, and $70 on Google, which gives 5,000 free search requests a month on Gemini 3 and newer then charges $14 per 1,000 and caps a project at 1,500 requests a day. None of those three returns the answer the ChatGPT app, Google AI Overviews or Copilot gave a logged-out user, and no model API returns a visibility figure.