AI engines
Best APIs to track Claude answers in 2026, led by API for GEO
Thirteen APIs cover Claude: seven name Anthropic's API as the method, two capture the interface, four publish none. Published rates run $2.94 to $160 per 1,000.

By Mathias C.SEO consultant, and more recently GEO, at MentionBox
Twenty-eight of the 55 AI visibility APIs we checked on October 23, 2026 return Claude answers, and the thirteen worth shortlisting split on one question: how they reach Claude. Seven of them name Anthropic's Messages API as the collection method, two describe capturing an interface a user sees, and four publish nothing on the subject. The published rate for 1,000 Claude answers runs from $2.94 with Rankability to $160 with Ahrefs Brand Radar, whose Claude module bills 8 checks per prompt instead of 1. API for GEO is our pick after hands-on testing by Mathias C., at $13 per 1,000 answers with share of voice, position and sentiment computed per brand.
This guide is for a developer or a small team that has to produce a weekly Claude figure and keep it defensible. It covers what a Claude answer actually is before you buy anything, why nearly every vendor goes through Anthropic's API, a section per API in the order of our general ranking, the real bill if you build it on the Claude API yourself, and the month-by-month cost of one realistic schedule. MentionBox, which publishes this site, has a commercial link with API for GEO.
The shortlist at a glance
| API | How it collects Claude | Claude model named | Price per 1,000 Claude answers | Metrics returned | Our score |
|---|---|---|---|---|---|
| API for GEO | not published; "executed natively on each engine" | no | $13, down to $7 at volume | share of voice, position, sentiment, themes, per-brand perception | 9.0 |
| LLM Pulse | user-visible responses; no per-engine detail | no | €30.67, plus a Claude add-on from €8 a month | share of voice, position, sentiment | 8.8 |
| DataForSEO | official Anthropic API, web_search optional | yes, you choose it | token cost passed through; $0.55051 in its own live example | answer, cited URLs, mentions, fan-out queries | 7.7 |
| AthenaHQ | not published | no | $81.94 | share of voice, position, sentiment | 7.4 |
| Promptwatch | official API "with a web search tool attached" | no | $15.83 | position, sentiment, source types | 7.4 |
| Rankability | not published | no | $2.94 | position, sentiment | 7.4 |
| Scrunch | browser automation or official APIs, chosen per platform | no | not published | share of voice, position, sentiment | 7.0 |
| Trakkr | official Anthropic API, no live web | no | not published | position, sentiment, no answer text | 6.8 |
| OtterlyAI | official API with web search | yes, Haiku | $15.75, plus a $109 Claude add-on | share of voice, position, sentiment | 6.7 |
| Peec AI | official Anthropic API | yes, Sonnet or Haiku | not published | share of voice, position, sentiment | 6.5 |
| Profound | "directly from the browser" | no | not published | share of voice, position, sentiment | 6.5 |
| Rankscale | official Anthropic API | yes, Haiku 5.5 and Sonnet 5.5 | $19.08 | position, sentiment | 6.5 |
| Ahrefs Brand Radar | official API, Claude Sonnet 4.6 with web search | yes, Sonnet 4.6 | $160, at 8 checks per Claude prompt | share of voice, sentiment, fan-out queries | 6.0 |
Published by each vendor on its pricing page, help centre and API reference, checked on October 23, 2026. Scores are ours, from those published facts, in the general ranking of 55 APIs. For API for GEO, four engines, plus the webhooks, the OpenAPI spec and the official SDKs, were confirmed in our hands-on test on October 26, 2026 and are not yet documented on apiforgeo.com.
Three columns in that table carry most of the decision. The method column tells you what population the figure describes. The model column tells you whether the figure survives next quarter's model release. The price column only compares once you have written down how many Claude answers a month you intend to buy.
What "a Claude answer" is, before you buy anything
Claude is two products that answer differently, and the gap between them is wider than for ChatGPT.
The Claude app searches the web on its own judgement. Anthropic's announcement, published on March 20, 2025 and updated on May 27, 2025, states that "web search is now available globally on all Claude plans" and that "when Claude incorporates information from the web into its responses, it provides direct citations so you can easily fact check sources". Its help article adds that in the current app "there's no web search toggle" and that Claude searches when searching helps. No API returns those answers.
The Claude API answers from the model you configure. To let it read the web you attach a web search tool to the request, which Anthropic's web search guide documents in three versions, web_search_20250305 for basic search and the two later ones for dynamic filtering and response inclusion. The guide is explicit that Claude decides: it searches when the request depends on information "that is current, changing, or outside its training data" and answers directly on "established facts, math, science fundamentals, or coding concepts". Triggering "is steerable through your system prompt", and max_uses caps the number of searches in a turn. There is no equivalent of a forced-search flag in the tool definition itself, which is the practical difference with OpenAI's tool, covered in our comparison of the OpenAI web search tool and ChatGPT answers.
What comes back is well shaped for measurement. Each citation block carries url, title, encrypted_index and up to 150 characters of cited_text, each search result carries url, title, page_age and an encrypted_content blob you must pass back verbatim in multi-turn conversations, and usage.server_tool_use.web_search_requests counts the searches you are billed for. Those fields are the citation layer of every Claude metric you will ever report.
Why nearly every vendor reaches Claude through Anthropic's API
For ChatGPT, several vendors run a real browser session against the interface. For Claude, almost nobody does, and the reason is written in the terms. Anthropic's consumer terms, effective October 8, 2025, forbid at section 3 using the service "to access the Services through automated or non-human means, whether through a bot, script, or otherwise" except when accessing through an Anthropic API key or with explicit permission, and separately forbid crawling, scraping or harvesting data from the services.
So the thirteen APIs here sort into three groups:
- Seven name the official API. DataForSEO, Promptwatch, OtterlyAI, Peec AI, Rankscale, Trakkr and Ahrefs Brand Radar all document Claude as an API model, four of them with a web search tool attached.
- Two describe a user-visible surface. Profound states that "we capture directly from the browser" and that it focuses "on front-end user experiences rather than API calls". LLM Pulse's help centre describes collecting user-visible responses without naming a method per engine.
- Four publish nothing per engine. API for GEO, AthenaHQ, Rankability and Scrunch. Scrunch is the most explicit about being unspecific, stating that it uses "a combination of browser automation and official platform APIs" and that "the collection method is selected per platform".
This is the question to put in writing before signing. A Claude share of voice built from Haiku answers with no web search, and one built from Sonnet answers with search forced on, are two different measurements that will land in the same slide.
1. API for GEO, Claude with the metrics already computed
API for GEO lists Claude among the nine engines on its AI Model Comparison page, which returns visibility, shareOfVoice, avgRanking, citationRate and totalResponses per model, so a Claude figure arrives next to the ChatGPT and Perplexity figures from one call rather than from three jobs you join yourself. One credit buys one prompt, executed once, against one model, in one locale, and its pricing page sets that credit at $0.013 up to 50,000 a month, then $0.011, $0.009 and $0.007, with no upfront commitment, credits that never expire, and reads over stored runs unmetered.
Its visibility endpoint takes the Claude model id directly:
curl -X POST https://api.apiforgeo.com/v1/visibility/entities \
-H "Authorization: Bearer $API_FOR_GEO_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "best open source vector database for a small team",
"models": ["claude"],
"country": "US",
"bucketSize": "week"
}'
- What comes back: one row per entity with
visibility,shareOfVoice,avgPosition,mentionsand asentimentobject, plus ranked sources with their category, themes and per-brand perception, fan-out questions, shopping products and YouTube citations, all inside the same credit. - Engines and markets: the ten we track. ChatGPT, Perplexity, Google AI Mode, Claude, Copilot and Grok are listed on apiforgeo.com; Gemini, Google AI Overviews, DeepSeek and Meta AI were confirmed in our hands-on test of October 26, 2026. Its published country list covers 187 countries, each run stated to execute in the country and language of the prompt.
- Integration: 300 requests a minute and 10 concurrent executions on pay as you go, an MCP server, plus webhooks, an OpenAPI spec and official SDKs confirmed in our test. Resale and white label are named as an intended use in section 6 of its terms.
- Limits, said plainly: $13 per 1,000 is ten times what a scraping API charges for a raw answer; its Claude collection method is not published, only the statement that prompts run natively on each engine; the key arrives on request through a form rather than at sign-up; and four engines plus the tooling come from our test rather than from its own pages.
- For whom: a developer who wants the computed layer for Claude without writing a counting pipeline, and who values no monthly floor more than the lowest unit price. It scores 9.0 of 10 in our survey, the highest, and first in our ranking for developers.
To try it, request a key on apiforgeo.com: it comes with a sandbox project.
2. LLM Pulse, Claude as a paid add-on on eighteen engines
LLM Pulse has the widest engine list we checked, eighteen in total including Mistral Le Chat, Qwen, Baidu AI and Naver AI Briefing, and the strongest integration: SDKs in twelve languages, HMAC-signed webhooks on Scale and above, an OpenAPI 3.0 spec and an MCP server on every plan. Its security annex states that the application, its databases and its job queues run on dedicated servers rented from Hetzner in Germany.
For Claude specifically, two facts set the budget. The API key starts on the €299 Scale plan, which works out at €30.67 per 1,000 answers with weekly tracking on our arithmetic over its published allowance, and Claude is a paid add-on from €8 per model per month rather than an included engine. Its published method covers user-visible responses, with no detail per engine, so the Claude line carries the same provenance question as the rest.
- For whom: a team that sells in markets the other vendors ignore and wants Claude in the same payload. It scores 8.8 of 10 in our survey and second in our ranking of GEO metrics APIs.
3. DataForSEO, Claude with forced web search at token cost
DataForSEO is the cheapest route to raw AI answers at $1.20 per 1,000 on its standard queue, and that rate does not apply to Claude. Its LLM Scraper, the one that scrapes interfaces, supports ChatGPT and Gemini only; Claude arrives through the LLM Responses endpoint, which calls Anthropic's official API and passes the model's token cost through on top of the task fee. The documented live example totals $0.55051 for one task, of which $0.54991 is the AI token cost, so the model and the output length you choose, rather than DataForSEO's rate card, set your bill.
What the endpoint gives you in exchange is the finest control of this comparison:
-
model_nameis required, so the Claude version behind every answer is yours to record. -
web_searchdefaults to false, andforce_web_searchturns Claude's own judgement off, which no other vendor here exposes. -
web_search_country_iso_codesets where the search runs;max_output_tokensdefaults to 2,048 and caps at 4,096;user_promptstops at 500 characters. -
The response carries
itemswith annotations,search_resultswith title, URL, date and snippet,fan_out_queries, and the token counts behind the price. -
Limits: no share of voice, position or sentiment, a $50 minimum payment, and a Claude cost you have to model yourself. SDKs in Python, C#, TypeScript and Java, webhooks, up to 100 tasks per call and an MCP server make it the easiest to industrialise. The duel API for GEO vs DataForSEO sets the two payloads side by side.
-
For whom: a developer who wants Claude answers with the model and the search setting pinned, and who already has the counting code.
4. AthenaHQ, ten engines without a published method
AthenaHQ covers all ten engines we track with share of voice, position and sentiment per brand, and it is the clearest of this group on identity plumbing, documenting SAML 2.0, OIDC and a SCIM 2.0 provisioning endpoint on Enterprise. Its API reference describes running tracked prompts against external AI providers without naming interface or API per engine, so the Claude row has no stated provenance.
- Price: $81.94 per 1,000 answers, our arithmetic on the $295 Starter plan and its 3,600 credits at one credit per AI response, and API access on Starter is a paid add-on quoted in-app, so the true unit cost is higher and unpublished.
- Integration and limits: 60 requests a minute per organisation, a response-streaming job you start and poll, an OpenAPI spec, an MCP server with read and write tools, no webhooks and no SDK. No published country list.
- For whom: a team that needs ten engines with single sign-on on a modest answer volume.
5. Promptwatch, Claude and Claude Code with a search tool attached
Promptwatch is the cheapest route back to a near-complete payload, its API comes with every paid plan, and it is the most honest vendor here about per-engine method. Its collection documentation states that "a few API models, like Claude and the coding assistants, run with a web search tool attached", and its pricing page lists Claude Code alongside Claude, Codex and OpenCode, which is the only coding-assistant surface in this comparison.
One GET /responses/{id} returns content, model, visibilityScore, sentimentScore, toneOfVoice, mentionedOurBrand, competitorMentions, citationCount, a brandMentions array with per-brand position and sentiment, and a citations array with url, domain, rank, sourceType and contentType. Its response limits page publishes the volume formula outright, prompts times models times runs.
- Price and limits: $15.83 per 1,000, from the $95 Essential plan and its 6,000 responses, which cannot be topped up. Share of voice is the one headline metric missing. No webhooks and no SDK, only a Postman collection, and rate limits are shared across REST, MCP and every key in the organisation.
- For whom: a developer who wants per-brand sentiment and position on Claude, at a predictable monthly fee, and who tracks what coding assistants say as well.
6. Rankability, the cheapest published rate for Claude
Rankability publishes the lowest rate per answer of any API here that covers Claude, $2.94 per 1,000, our arithmetic on its $97 Starter plan, and it states that all three paid plans carry the same platform coverage, so Claude is not an add-on. All ten engines we track are covered, Brave AI as well, and it returns the answer, the cited URLs, brand mentions, position and sentiment, with an MCP server to read them.
- Limits: no share of voice, no published collection method for any engine, no SDK and no webhooks, and its terms forbid resale, which rules it out for an agency reporting under its own brand.
- For whom: a side project or an internal dashboard where the unit price decides and an analyst can compute the portfolio view.
7. Profound, the one vendor that says it captures Claude in a browser
Profound is the only API in this comparison whose published method for Claude is a user-visible surface, stating that it captures directly from the browser and that what you see in Profound is what your customers see. It also has the widest documented footprint, 150-plus regions and 30-plus languages. Its visibility report is a POST on /v2/reports/visibility returning visibility_score, share_of_voice and average_position, grouped by date, model, topic, region, prompt or persona, and its answers endpoint adds the response text, a citation_details array, a deduplicated mentions list and sentiment_claims by asset and theme.
- Access and limits: its API reference states the REST API is available on the Enterprise plan, with no published price, and the free trial runs 50 prompts daily for seven days on three engines without API access. Delivery is synchronous only at 600 requests an hour per key by default, and its Master Subscription Agreement bars distributing, selling or sublicensing the services. Python and TypeScript SDKs, an OpenAPI spec and a hosted MCP server are published.
- For whom: a brand team that needs Claude answers as logged-out users see them and can sign an enterprise contract. Our guide to Profound alternatives with a self-serve API covers the field at published prices, and API for GEO vs Profound goes line by line.
8. Peec AI, model channels that survive a Claude upgrade
Peec AI computes share of voice, position, sentiment and source categories over nine of our ten engines, lists Claude Sonnet and Claude Haiku as options, and publishes the formula behind each metric, which we would copy elsewhere: share of voice is mention_count / sum(mention_count), visibility is sum(visibility_count) / sum(visibility_total), position is sum(position_sum) / sum(position_count).
Its best idea solves the problem every Claude tracker hits twice a year. Its model channel concept defines a channel as a stable identifier for one AI surface that survives a model upgrade, so a year-long Claude series filtered by model_channel_id stays intact where filtering by model name fragments at every release. Build that identifier into your own schema whatever vendor you buy.
- Access and limits: its changelog records under v0.39.0, dated October 27, 2026, that API availability moved from Enterprise to Pro, Advanced, Enterprise and agency plans, with no price per answer published on any tier. 200 requests a minute per project, an OpenAPI spec, an MCP server on every paid plan, no SDK and no webhooks, no endpoint that executes a prompt on demand, and resale forbidden at 1.3 and 1.6 of its terms. Our face-off API for GEO vs Peec AI sets the two payloads side by side.
The rest of the field
| API | Claude-specific fact that decides it | Entry ticket |
|---|---|---|
| Scrunch | Collection method chosen per platform and not disclosed per engine; the Responses API carrying answer text is Enterprise only | Agency Core, $500 a month |
| Trakkr | Sends each prompt to the providers' own APIs, and its FAQ states only Perplexity and AI Overviews read the live web, so Claude answers come without live sources | Scale, $500 a month |
| OtterlyAI | Tracks "Claude via API, using Haiku model with web search capabilities", sold as a $109 a month add-on on the $189 Standard plan | $189 plus $109 for Claude |
| Rankscale | Claude sits in its model-engine list rather than its interface list; its changelog added Claude Haiku 5.5 and Sonnet 5.5 on October 12, 2026 | Pro+, $229 a month |
| Ahrefs Brand Radar | Claude Sonnet 4.6 with web search, custom prompts only, and each Claude check consumes 8 checks, which takes $20 per 1,000 answers to $160 | Lite, $129 a month, 2,500 checks for $50 |
Published by each vendor, checked on October 23, 2026. The $160 figure is our arithmetic on the published 8-checks rule and the $50 for 2,500 checks package.
Two of these deserve a second look despite their rank. OtterlyAI is the only vendor that names its Claude model and its search setting in the same sentence, which is exactly the disclosure the rest of the market owes its buyers. Ahrefs makes reads free, stating that pulling prompt data for custom queries by API consumes no API units, so the collection cost is the whole cost.
Build it yourself on the Claude API
For Claude, the build route is unusually competitive, because the official API is the method most vendors use anyway. The call is short:
curl https://api.anthropic.com/v1/messages \
-H "x-api-key: $ANTHROPIC_API_KEY" \
-H "anthropic-version: 2026-10-21" \
-H "content-type: application/json" \
-d '{
"model": "claude-haiku-5-5",
"max_tokens": 1024,
"messages": [
{"role": "user", "content": "best open source vector database for a small team"}
],
"tools": [{
"type": "web_search_20250305",
"name": "web_search",
"max_uses": 3,
"user_location": {
"type": "approximate",
"country": "US",
"city": "Chicago",
"timezone": "America/Chicago"
}
}]
}'
The answer arrives as content blocks: a server_tool_use block per search with the query Claude wrote, a web_search_tool_result block with the results, and text blocks whose citations array carries url, title and cited_text. Your counting job reads the text for brand mentions and the citations for your own domains, and our walkthrough of tracking AI citations of your pages with an API covers the matching rules.
Three published numbers set the bill. Web search is "$10 USD per 1,000 searches" plus the tokens of the search content, and a search that fails with an error is not billed. Anthropic's pricing page puts Claude Haiku 5.5 at $0.10 per million input tokens and $0.50 per million output for prompts up to 100,000 tokens, Claude Sonnet 5.5 at $2 and $10, and Claude Opus 5.5 at $4 and $20. The Batch API takes 50% off both token directions, while web search inside a batch is priced the same as in a normal request.
| Route for 1,000 Claude answers, one search each | Searches | Tokens | Total |
|---|---|---|---|
| Claude Haiku 5.5 | $10 | about $1.07 | about $11 |
| Claude Sonnet 5.5 | $10 | about $21.39 | about $31 |
| Claude Opus 5.5 | $10 | about $42.77 | about $53 |
Our arithmetic on Anthropic's published rates, read on November 11, 2026, applied to the usage reported in the example response of its web search guide: 6,039 input tokens and 931 output tokens for one searched answer. Your own token counts will differ with the prompt and the number of searches.
So a developer who writes the counting code pays about $11 per 1,000 Claude answers on Haiku, against $13 for API for GEO with every metric computed and stored. Two other published limits matter at volume: the Start tier allows 4,000 requests a minute on Claude Haiku 5.5 and carries a $500 monthly spend cap, and the Batches API throttles web search per organisation, so a large overnight batch with a search on every request can take longer than the batch itself suggests.
What one month of Claude tracking costs
Take a developer-sized schedule: 50 prompts, three runs a week, Claude only, one country. That is 50 × 3 × 52 ÷ 12, about 650 Claude answers a month.
| Route | Monthly cost for about 650 Claude answers |
|---|---|
| Build it on the Claude API, Haiku 5.5 | about $7.20, and you write the counting |
| API for GEO | about $8.45 at $0.013 a credit, metrics included, no plan floor |
| Rankability | $97, the Starter floor; 650 answers use about $1.91 of it |
| Promptwatch | $95, the Essential floor; 650 of its 6,000 responses |
| Ahrefs Brand Radar | about $104 of checks at 8 checks per Claude prompt, plus the $129 Lite plan |
| Rankscale | $229, the Pro+ floor |
| OtterlyAI | $298: the $189 Standard plan plus the $109 Claude add-on |
| AthenaHQ | $295 Starter plus a paid API add-on quoted in-app |
| LLM Pulse | €299 Scale plus the Claude add-on from €8 a model |
| Scrunch, Trakkr | $500 a month entry plan, no published volume |
| Peec AI, Profound | on quote |
| DataForSEO | token cost of your chosen model, after a $50 minimum payment |
Our arithmetic on each vendor's published rate and plan allowance, prices checked on October 23, 2026, and on Anthropic's published rates for the first line. Plan floors are included where the key sits behind a plan.
At this volume the monthly fee decides, not the unit price, and the two routes without a floor win by a factor of ten. The ranking inverts at scale: our note on the cost of tracking 100 prompts on five AI engines shows what happens once the answer count passes a plan's allowance.
Five mistakes we see on a Claude shortlist
- Taking "Claude" as one engine. Haiku with no search, Sonnet with search, and the app a buyer actually uses give three different answers to the same prompt. Record the model identifier and the search count next to every figure, and keep a stable channel id as Peec AI does.
- Comparing a Claude mention rate to a ChatGPT one. Several vendors scrape the ChatGPT interface and call the Claude API, so one engine's row is a user-visible answer and the other is a configured model. Our explainer on share of voice, visibility and position shows how far two defensible definitions can diverge on the same answers.
- Forgetting that Claude may not search at all. An answer with no citations is still an answer, and it tells you what the model knows rather than which pages it reads. Count the two populations separately.
- Buying an add-on rate as if it were the plan rate. Claude is a paid extra on OtterlyAI and LLM Pulse, and it costs 8 checks instead of 1 on Ahrefs Brand Radar. Price the Claude line, not the platform line.
- Running one prompt once a month. One run is one draw from a distribution. Three to five runs per prompt per period is the floor for a trend you can defend, and our guide to building a prompt set for GEO covers how many prompts you need before that.
Our view
Three defensible answers, depending on which constraint binds.
- You want computed Claude metrics and no monthly floor. API for GEO is our pick: Claude listed among its engines, share of voice, position, sentiment, themes and per-brand perception in one response, $13 per 1,000 answers down to $7, unmetered reads over stored runs, resale allowed in writing. The trade-offs to accept are the price per answer, the key on request, and a Claude collection method it does not publish.
- You want full control of the model and the search setting. Build it on Anthropic's Messages API at about $11 per 1,000 answers on Claude Haiku 5.5, or buy DataForSEO's LLM Responses endpoint for Claude and use
model_name,force_web_searchandweb_search_country_iso_codeto pin every variable. Both leave the counting to you. - You need Claude answers as users see them. Profound is the only vendor here that publishes browser capture for Claude, in 150-plus regions, on an Enterprise quote and with resale forbidden.
Whichever route you pick, write down three things before the first call: the Claude answers you need per month, the model and search setting you want behind them, and whether you or the vendor computes share of voice. Our ranking for developers scores the self-serve options on price, tooling and the data they return, and our GEO metrics ranking keeps only the APIs that compute the metrics for you.
Frequently asked questions
Is there an official API that returns Claude app answers?
No. Anthropic publishes the Messages API, where you attach a `web_search` tool to the request, and that answer comes from the model and the settings you chose rather than from the Claude app. Web search costs $10 per 1,000 searches plus the tokens of the search content, and each search counts as one use whatever the number of results. Anthropic's consumer terms, effective October 8, 2025, also forbid accessing the service "through automated or non-human means, whether through a bot, script, or otherwise" except with an API key, which is why interface capture is rare for Claude.
Which API returns share of voice and sentiment for Claude?
API for GEO, our editorial pick after hands-on testing by Mathias C., plus LLM Pulse, AthenaHQ, Scrunch, OtterlyAI, Peec AI and Profound. API for GEO returns share of voice, position, sentiment, ranked sources, themes and per-brand perception in one response for one credit, at $13 per 1,000 answers down to $7 at its largest tier with no monthly minimum, and reads over stored runs are not billed. DataForSEO, Rankability, Promptwatch and Rankscale return the answer, the cited URLs and the brands named, and leave share of voice to you. Prices checked on October 23, 2026.
Does Claude search the web on every prompt?
No, and that decides what you are measuring. Anthropic's web search guide states that Claude searches when the request depends on information that is current or outside its training data, and answers directly from stable knowledge otherwise, with the `max_uses` parameter as the only hard cap. There is no parameter that forces a search, so vendors steer it: DataForSEO exposes `web_search` with a default of false plus a `force_web_search` flag, Promptwatch documents that Claude runs "with a web search tool attached", and Trakkr's FAQ states that only Perplexity and AI Overviews read the live web on its platform.
Which Claude model do these APIs query?
It differs by vendor, and the model changes the answer. OtterlyAI states it tracks "Claude via API, using Haiku model with web search capabilities", Ahrefs Brand Radar names Claude Sonnet 4.6 with web search, Peec AI lists Claude Sonnet and Claude Haiku as options, and Rankscale's changelog added Claude Haiku 5.5 and Sonnet 5.5 on October 12, 2026. Ask every vendor to write the model identifier next to each figure, because a token price gap of twenty times sits between Claude Haiku 5.5 at $0.10 per million input tokens and Claude Sonnet 5.5 at $2.
How much does it cost to track 50 prompts in Claude every week?
Fifty prompts run three times a week is about 650 Claude answers a month. At the rates published on October 23, 2026 that is about $8.45 with API for GEO, metrics included, or $1.91 of usage with Rankability inside its $97 Starter plan. Built on the Claude API with one search per answer it costs about $7.20 on Claude Haiku 5.5 and about $20.40 on Claude Sonnet 5.5, counting $10 per 1,000 searches and the token usage of the example response in Anthropic's documentation, and you write the counting code.
Can I resell Claude visibility data to clients?
Check the terms per vendor, because the split is wide. API for GEO names resale and white label as an intended use in section 6 of its terms, LLM Pulse allows branded client use on eligible plans, and AthenaHQ, Promptwatch, Rankscale and Trakkr allow it. Rankability, Peec AI, Profound and Ahrefs Brand Radar do not: Profound's Master Subscription Agreement bars distributing, selling or sublicensing the services, and Peec AI's terms state at 1.3 that Peec is not appointing the customer as a reseller.