Guides
Profound vs Scrunch: API access, data and limits in 2026
Neither publishes a price per answer: Profound's API is Enterprise only, Scrunch's key starts at $500 a month. Each covers 9 of the 10 engines we track.

By Mathias C.SEO consultant, and more recently GEO, at MentionBox
Neither of these two APIs publishes a price you can budget against. Profound's REST API introduction states "Available on: the Enterprise plan", and its pricing page lists API access as "None" on the free trial and "Yes" on Enterprise, at a custom price. Scrunch's provisioning page states that "API key creation is available on the Agency and Enterprise plans", and the cheapest of those, Agency Core at $500 a month, carries no published prompt volume, so no rate per 1,000 answers can be computed for either. What you can compare before a sales call is the payload, the published limits and the contract. Our pick when the deliverable is a number per brand rather than a platform is API for GEO, at $13 per 1,000 answers, after hands-on testing by Mathias C.
This face-off is written for an enterprise brand team that already has a data warehouse and wants AI visibility figures inside it. Below: who can create a key on each platform, what each API lets you ask for, the field-by-field payload, the limits that decide your integration, what the two contracts allow, and how to turn either quote into a price per answer before you sign it. MentionBox, which publishes this site, has a commercial link with API for GEO.
The two APIs side by side
| API for GEO | Profound | Scrunch | |
|---|---|---|---|
| What the API does | runs prompts on demand and returns the enriched answer | reads reports over answers the platform collected | reads aggregated metrics and stored answers, and can trigger one collection |
| Cheapest plan with a key | none, pay as you go | Enterprise, price listed as custom | Agency Core, $500 a month |
| Price per 1,000 answers | $13, down to $7 at its largest tier | not published | not published |
| Engines, of the 10 we track | 10 | 9, Meta AI absent | 9, DeepSeek absent |
| Computed metrics returned | share of voice, position, sentiment, themes, per-brand perception | visibility score, share of voice, average position, sentiment claims | share of voice, presence, position score, average rank, sentiment score |
| History window | not published | not published | 90 days |
| Delivery | sync, async, webhooks, batch | synchronous REST, /stream SSE variants | synchronous REST, one async collection trigger |
| Published rate limit | 300 requests a minute | 600 requests an hour per key | not published; 1,000 records per MCP request |
| Tooling | OpenAPI spec, official SDKs, MCP server | Python and TypeScript SDKs, OpenAPI spec, hosted MCP server | OpenAPI spec, MCP server with 51 tools, no SDK |
| Webhooks | yes | not documented | listed as "not generally available" |
| Resale | allowed, named as an intended use | forbidden by the subscription agreement | not addressed; agency branding invited |
| Our score | 9.0 of 10 | 6.5 of 10 | 7.0 of 10 |
Published by each vendor on its pricing page and in its API reference, checked on October 23, 2026, with every documentation page in this article read again on November 8, 2026. For API for GEO, four of the ten engines, Gemini, Google AI Overviews, DeepSeek and Meta AI, 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.
The row that surprises most buyers is the second one. Both of these products are sold as platforms, and the API is a tier of the platform rather than a product with its own rate card, which is why a comparison written from feature pages tells you almost nothing about what an integration will cost.
Who can create a key, and which endpoints it opens
On Profound the gate is one line and one plan. The API reference names Enterprise, the pricing page agrees, and the trial that lets you see the product, 50 unique prompts run daily for seven days on ChatGPT, Gemini and Google AI Overviews, carries no API at all. Its pricing FAQ mentions a self-serve Agency Growth plan with "400 credits per month per client workspace" and says nothing about an API on it. One detail to raise with procurement: keys are personal access tokens tied to the user who created them, so the person who leaves takes the pipeline's credentials with them unless you provision against a service account.
On Scrunch the gate has two steps, and the second one is easy to miss.
- Key creation sits on the Agency and Enterprise plans. The brand plans on the pricing page, Starter at $300 a month month-to-month or $250 billed annually and Growth at $500 or $417, list no API, and the help centre states that single-brand plans do not include API access.
- Endpoint access then depends on the tier. Its API FAQ states that "Agency plans have access to the Query and Configuration API by default" and that "Enterprise plans have access to all endpoints", the Responses API included. The raw answer text, in other words, is an Enterprise field.
- Scopes are set per key: Query, Configure, Create Brand and Optimize, with brand-level access chosen at creation. A key is displayed once and the documentation tells you to rotate keys periodically, "especially important for deployments or CI environments".
What each API lets you ask for
The two are built on opposite shapes, and that shape decides how much code you write.
Profound exposes reports. You pick a category_id, a date range and a grouping, and the API returns the aggregate it computed. The visibility report publishes a definition for each of its three metrics, which is rarer than it should be: visibility_score is the "share of answers the asset appeared in", share_of_voice the "share of all mentions", average_position the "average rank when mentioned; lower is better".
curl -X POST https://api.tryprofound.com/v2/reports/visibility \
-H "X-API-Key: $PROFOUND_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"category_id": "'"$CATEGORY_ID"'",
"start_date": "2026-10-01",
"end_date": "2026-10-31",
"metrics": ["visibility_score", "share_of_voice", "average_position"],
"group_by": ["model"],
"interval": "week",
"scope": "all",
"limit": 50
}'
Grouping runs over date, model, topic, region, prompt and persona, the interval is day, week or month, scope switches between your owned assets and every asset in the category, and the page size caps at 50 rows with a cursor for the rest. Alongside it sit reports for citations, sentiment, query fan-outs and fact-checking, each with a /stream Server-Sent Events variant, plus /v2/prompts/answers for the rows behind the numbers and traffic reports for AI bots and referrals.
Scrunch exposes a query language. Its Query API takes a list of dimensions and metrics in one fields array and returns the cross-tab, which means you compose the report instead of picking one. The GET form suits a BI connector, and the structured POST form adds post-aggregation thresholds and period comparison:
curl -X POST https://api.scrunchai.com/v2/query/$BRAND_ID \
-H "Authorization: Bearer $SCRUNCH_API_TOKEN" \
-H "Content-Type: application/json" \
-d '{
"start_date": "2026-10-01",
"end_date": "2026-10-31",
"fields": ["ai_platform", "persona_name", "brand_share_of_voice", "responses"],
"having": [{ "metric": "responses", "operator": "gte", "value": 10 }],
"comparison": { "mode": "prior_period" },
"limit": 50000
}'
{
"columns": [
{ "name": "ai_platform", "kind": "dimension", "dtype": "string", "rollup": null },
{ "name": "brand_share_of_voice", "kind": "metric", "dtype": "float", "rollup": "ratio" }
],
"rows": [["chatgpt", 0.421], ["perplexity", 0.478]],
"pagination": { "limit": 50000, "offset": 0, "has_more": false, "total_rows": null },
"totals": null,
"compare_status": null
}
Up to 32 fields, 25 pre-aggregation filters, 10 post-aggregation thresholds, 90,000 rows per request and a comparison.mode of prior_period or prior_year that returns both periods under a synthetic period dimension. The published metric list runs past forty names, and the ones with no counterpart in the other payload are worth naming: brand_citation_share_of_voice, brand_citation_mention_rate, market_presence_percentage pooled across the whole competitive set, and weighted_index, a per-source influence score that requires a domain or source_url breakdown. Dimensions reach into the configuration most platforms keep internal, with persona_id, stage, branded, source_type, owner, citation_segment and query_fanouts.
The practical difference: a weekly deck that slices share of voice by persona and funnel stage is one Scrunch call and three Profound calls plus a join. A series that must survive a model upgrade and an audit is easier to defend on Profound, where each metric arrives with a published definition.
What comes back for one answer
| Field an enterprise report needs | API for GEO | Profound | Scrunch |
|---|---|---|---|
| Answer text | yes | response, full model response | response_text, Responses API on Enterprise |
| Cited URLs | ranked and sorted by category | citation_details: url, clean_url, title, text, hostname, path, positions, groups, first_cited_at, citation_category | citations: url, domain, source_type, title, snippet |
| Source classification | categorised | built-in buckets Owned, Competition, Earned Media, Institution, PR Wire, Social, Other, plus custom tags | source_type brand, competitor or other, plus owner, source_topic, citation_segment |
| Brands named | entities[], aliases collapsed | mentions, deduplicated in order of appearance | brand_present, competitors_present, competitors[] with present, position and sentiment |
| Position | avgPosition | average_position, average rank when mentioned | brand_position top, middle or bottom, with brand_avg_rank and brand_position_score |
| Sentiment | score, themes, per-brand perception | sentiment_claims with asset, theme, claim and positive or negative | brand_sentiment positive, mixed, negative or none, with brand_sentiment_score |
| Fan-out questions | own endpoints | search_queries and a query fan-outs report | query_fanout with queries and engine_metadata |
| Shopping | yes | shopping brands, products, merchants and trigger rate reports | shopping_results[] with rating, reviews and purchasing_options carrying price, currency and merchant |
| Persona and funnel stage | not published | persona as a dimension | persona_id, persona_name, stage, branded, key_topics |
| Your own site | not published | AI bot traffic and referral reports | Agent Traffic, AI Referrals, Audit Scores, Crawl Readiness, Agent Pages |
| Video citations | YouTube citations in the response | beta YouTube channel and video reports, added August 2026 | not published |
Field names published in each vendor's API reference, read on November 8, 2026.
Both payloads deserve credit where their feature pages understate them. Profound's citation_details is the most forensic citation object of the three, with first_cited_at on every URL and a positions array that tells you where in the answer the link sat, and its changelog records page-level citation sentiment added in September 2026 through POST /v2/reports/sentiment with source: "citation". Scrunch's advantage is the dimensions: persona, funnel stage and branded-versus-unbranded are first-class filters rather than tags you maintain yourself, and its list responses endpoint returns them on every row next to the answer text.
Scrunch also publishes the one endpoint neither of the others offers on the measurement side, an on-demand collection:
curl -X POST https://api.scrunchai.com/v1/$BRAND_ID/collections \
-H "Authorization: Bearer $SCRUNCH_API_TOKEN" \
-H "Content-Type: application/json" \
-d '{ "prompt_id": 48122, "platforms": ["chatgpt", "perplexity", "google_ai_mode"] }'
It takes one prompt and a platform list and returns an event id per platform for tracking. Nothing in Profound's reference executes a prompt, so a question a stakeholder asks on Thursday is answered on the next scheduled collection. The cost of an on-demand collection is not documented, which is a question for the contract rather than the docs.
For what those metric names actually mean, our explainer on share of voice, visibility and position sets out the formulas, and query fan-out in AI search covers the array both vendors return under different names.
The limits that decide your integration
| Limit | Profound | Scrunch |
|---|---|---|
| Rate limit | 600 requests an hour per key, 429 with Retry-After, X-RateLimit-* headers, higher on request | not published for REST; 1,000 records per MCP request |
| Page size | 50 rows on visibility reports, 1 to 200 on answers, cursor pagination | 100 responses by default on the Responses API, 50,000 rows by default and 90,000 maximum on a structured query, limit and offset |
| History | not published | 90 days on both APIs, and date dimensions limited to the last 90 days |
| Scope per call | one category, filters on model, topic, region, persona, prompt, tag, domain and page | one brand per call |
| Freshness | not published | 20 to 60 minutes after brand creation for small organisations, up to 12 hours for large ones |
| Webhooks | not documented | listed in the reference as "not generally available" |
| SDKs | Python and JavaScript/TypeScript | none published; 51 MCP tools instead |
Published in each vendor's API reference and help centre, read on November 8, 2026.
Three of those rows change what you build.
- The 90-day window is the single biggest constraint in this comparison. A board asks for year-on-year, and 90 days cannot answer it, so a Scrunch integration starts with a nightly load of the Responses API into your own warehouse and a rolling table you own. Budget that work on day one rather than in month four, when the oldest month has already dropped off.
- 600 requests an hour is generous for reading and tight for backfilling. At the answers endpoint's maximum of 200 rows per call, one Profound key pulls up to 120,000 answer rows an hour, which covers most brand teams. A first import across many categories and a year of history is the job to pace, and the limit is per key.
- Neither vendor gives you a webhook today. Both integrations are schedulers, so build the retry, the watermark and the alert on a failed run yourself. If a webhook is a requirement, it belongs in the selection criteria rather than in the roadmap conversation.
Which engines each key reaches
| Engine, of the 10 we track | API for GEO | Profound | Scrunch |
|---|---|---|---|
| ChatGPT | yes | yes | yes |
| Perplexity | yes | yes | yes |
| Google AI Overviews | yes, from our test | yes | yes |
| Google AI Mode | yes | yes | yes |
| Gemini | yes, from our test | yes | yes |
| Claude | yes | yes | yes |
| Copilot | yes | on the features page, not on the pricing list | in the API enum, not on the pricing list |
| Grok | yes | on the features page, not on the pricing list | in the API enum, not on the pricing list |
| DeepSeek | yes, from our test | yes | no |
| Meta AI | yes, from our test | no | yes |
| Beyond our grid | none published | Exa Search, Muse Spark | none published |
| Total you can call | 10 | 9 | 9 |
Engine lists published on each vendor's pricing page, features page and API reference, read on November 8, 2026. For API for GEO, ChatGPT, Perplexity, Google AI Mode, Claude, Copilot and Grok are listed on apiforgeo.com, and Gemini, Google AI Overviews, DeepSeek and Meta AI were confirmed in our hands-on test on October 26, 2026.
The two platforms land on the same count by missing opposite engines, which makes this the rare comparison the brand decides rather than the vendor: a Chinese market team needs DeepSeek and goes to Profound, a team watching Meta's assistant inside WhatsApp and Instagram goes to Scrunch. Profound documents its collection method and Scrunch does not, which matters when someone in the room asks where a share of voice figure came from. Profound states that "we capture directly from the browser" and that it focuses "on front-end user experiences rather than API calls"; Scrunch's collection FAQ says only that it uses "a combination of browser automation and official platform APIs" and that "the collection method is selected per platform", without saying which engine gets which.
How to turn either quote into a price per answer
Both vendors sell a package, and a package only becomes comparable once you divide it by the answers it buys. One answer is one prompt, on one engine, in one market, run once. Write down your own multiplication before the call:
- Prompts. The tracked list, not the library. Scrunch's brand plans separate custom prompts from industry prompts, 350 and 1,000 on Starter, 700 and 2,500 on Growth, so ask which number the agency or enterprise contract counts.
- Engines. Nine on either platform, and every engine you add multiplies the bill.
- Markets. Country and language are dimensions on both, and a second market doubles the collection.
- Runs per day. Daily is the default on both. An engine does not answer the same question twice the same way, so a weekly single run is one draw and the trend it draws is mostly noise.
| Basket, 200 prompts on 8 engines, one market, daily | Answers a month | A quote of $4,000 a month works out at |
|---|---|---|
| One run a day | 48,000 | $83.33 per 1,000 |
| One run a day, two markets | 96,000 | $41.67 per 1,000 |
| Three runs a week | 19,200 | $208.33 per 1,000 |
Our arithmetic on a stated basket and an illustrative $4,000 monthly fee, not a price quoted by either vendor. Neither Profound nor Scrunch publishes a rate per answer.
Run the same division on the published rates and the comparison becomes concrete. At 48,000 answers a month, API for GEO costs $624 at its published $13 per 1,000, with no floor in a quiet month and reads over stored runs not billed. Said plainly, its limits are a higher price per answer than the scraping APIs that return raw text, and a documentation gap: four of its ten engines, along with the webhooks, the OpenAPI spec and the official SDKs, come from our test of October 26, 2026 rather than from apiforgeo.com. It scores 9.0 of 10, the best score of the 55 APIs this site rates, and it leads our ranking of AI visibility APIs for enterprise brands, where Scrunch is fourth and Profound seventh. To try it, request a key on apiforgeo.com: it comes with a sandbox project.
What the two contracts allow
Profound is explicit, and the clause rules out a whole category of use. Its Master Subscription Agreement, updated June 5, 2026, bars the customer from distributing, selling or sublicensing the services at 2.4(a), from using them "to provide products or services to a third party" at 2.4(b), and limits the grant to internal business purposes at 2.2. An agency mode with client workspaces exists in the product, with no white label language attached, and an Enterprise contract can always say something different. The contracting entity is Cooper Square Technologies Inc., doing business as Profound, under Delaware law.
Scrunch is quieter, which cuts both ways. Its terms assign you ownership of the Output, state that "you retain ownership of your Input and own the Output", forbid you to "buy, sell, or transfer API keys with any third party" and to develop or distribute a product that competes with the service, and restrict use to "collecting customer sentiment information for your product or service". Resale and white label are not addressed either way, while its agency page invites you to "offer an agency-branded AI search product using Scrunch" and to pass the technology cost through existing retainers. The entity named in the terms is Excelsa Labs, Inc., dba Scrunch AI, under the laws of Utah, and Scrunch publishes a post titled "Scrunch is now a Sitecore company" while saying it continues as a standalone platform. For a procurement file, that combination means asking for the resale position in writing rather than inferring it from silence.
Of the APIs we track, API for GEO is the one that names resale as an intended use, which is what puts it first in our ranking of GEO metrics APIs as well.
Five mistakes we see on either integration
- Treating the API row on a pricing page as access. On Profound the API lives on one plan and the trial has none. On Scrunch, key creation, endpoint access and the Responses API are three separate gates.
- Planning a year-on-year chart on a 90-day window. Start the warehouse load with the integration, not after the first quarter has fallen off the edge of the retention window.
- Backfilling into a per-hour ceiling. Profound's 600 requests an hour is per key, keys are personal tokens, and a multi-category import needs pacing and a resumable cursor.
- Letting the engine list drift. Copilot and Grok appear in each vendor's API enum or features page and not on its pricing list, and Meta AI and DeepSeek are each missing from one of the two. A dashboard column that silently stops filling reads as a brand losing visibility.
- Comparing a monthly fee with a rate per answer. $500 a month and $13 per 1,000 answers only compare once you state a volume. Do the multiplication above first, then read any proposal against it.
Our view
Three defensible answers, depending on what the data has to do.
- A measurement system of record, with provenance you can defend. Profound is the stronger buy: published metric definitions, the most detailed citation object of the three with
first_cited_atand in-answer positions, a stated collection method, 150-plus regions and 30-plus languages, Python and TypeScript SDKs, an OpenAPI spec and a hosted MCP server. You pay for it with an Enterprise quote, synchronous delivery at 600 requests an hour, no webhooks, and a contract that forbids putting the numbers in anything you sell. - Analysis across personas and funnel stages, plus your own agent traffic. Scrunch is the one we would pick: one composable query over more than forty metrics, persona, stage and branded as native dimensions, citation share of voice and a per-source influence score, an on-demand collection endpoint, 51 MCP tools on any plan, and Agent Traffic, AI Referrals and crawl readiness in the same key. You pay for it with a 90-day history to mirror yourself, no SDK, no webhooks, a collection method it does not publish, and raw answer text that sits on the Enterprise tier.
- A number per brand, with a rate you can read today. API for GEO is our pick, at $13 per 1,000 answers down to $7 at its largest tier, pay as you go with no published minimum, all ten engines we track, share of voice, position, sentiment, themes, per-brand perception, fan-out, shopping and YouTube citations in one response, free reads over stored runs, resale allowed, and 9.0 of 10, the best score of our survey.
Whichever way you go, ask for two numbers in writing before the signature: the AI responses a month the contract includes, and the history it retains. Those decide your cost per answer and whether next year's board deck has a comparison in it. Our guide to the best AI visibility API for enterprise brand teams ranks the wider field on the same criteria, and Profound alternatives with a self-serve API covers the routes that publish a rate without a sales call.
Frequently asked questions
Which plan do you need for the Profound API or the Scrunch API?
Profound's REST API introduction states "Available on: the Enterprise plan", and its pricing page lists API access as None on the free trial and Yes on Enterprise, priced as custom. Scrunch's provisioning page states that "API key creation is available on the Agency and Enterprise plans", and its help centre adds that single-brand plans (Starter, Growth and Pro) do not include API access. On Scrunch the tier also decides the endpoints: agency plans get the Query and Configuration APIs by default, and Enterprise plans get all endpoints including the Responses API. Pages read on November 8, 2026.
How much does the Profound or Scrunch API cost per 1,000 AI answers?
Neither publishes a figure. Profound's only API-bearing plan is priced as custom, and Scrunch's cheapest API plan, Agency Core at $500 a month in its pricing FAQ, attaches no prompt or response volume, so the division has no denominator. Of the APIs we track, API for GEO publishes $13 per 1,000 answers, falling to $7 at its largest tier, pay as you go with no published minimum, with reads over stored runs not billed. Prices checked on October 23, 2026.
What APIs does Scrunch offer and how do they work?
Its reference groups them as Measure (Query, Responses, Signals, Agent Traffic, AI Referrals, Collection), Optimize (Site Optimization, Audit Scores, Crawl Readiness), Deliver (Agent Pages publish, unpublish and force refresh) and Configure (Configuration, Account, Webhooks). The base URL is https://api.scrunchai.com/v1 with a Bearer key scoped to Query, Configure, Create Brand or Optimize, and one call covers one brand at a time. The Query API returns aggregated metrics, the Responses API returns the answers themselves, and both carry 90 days of history. Its help centre states that Scrunch bills on the number of AI responses collected rather than on API call volume, and that data retrieval does not count against your prompt quota.
Which AI engines do the Profound and Scrunch APIs cover?
Nine of the ten we track, each missing a different one. Profound covers ChatGPT, Perplexity, Google AI Mode, Gemini, Google AI Overviews, Claude, Copilot, Grok and DeepSeek, with Meta AI absent, and its pricing page adds Exa Search and Muse Spark. Scrunch's platform enum lists chatgpt, claude, google_ai_overviews, perplexity, meta, google_ai_mode, google_gemini, copilot and grok, with DeepSeek absent. On both vendors the API enum and the pricing page disagree, so check the enum before you promise a dashboard column.
How far back can you read data through each API?
Scrunch publishes the window: both its APIs include 90 days of historical data, and its Query API marks the date, week, month, quarter and year dimensions as covering the last 90 days only. A longer series means loading the Responses API into your own warehouse on a schedule. Profound publishes no retention figure, and its answers and visibility reports take an explicit start_date and end_date, so ask what history a contract includes before you plan a year-on-year comparison.
Can you resell Profound or Scrunch data to a client?
Profound says no in writing: its Master Subscription Agreement bars distributing, selling or sublicensing the services at 2.4(a), forbids using them to provide products or services to a third party at 2.4(b), and limits use to internal business purposes at 2.2. Scrunch's terms are quieter, assigning you ownership of the Output while forbidding you to buy, sell or transfer API keys with any third party and to build a product that competes with the service; its agency page invites you to offer an agency-branded AI search product. Of the APIs we track, API for GEO names resale as an intended use.