Recomma vs Scrunch (2026): agent traffic and AXP, or measured answers

Scrunch and Recomma both ask AI assistants your buyers' questions and report which brands the answers name. Scrunch reaches further into the website itself: it watches which AI agents visit, and its Agent Experience Platform serves those agents a stripped-down, machine-readable version of each page. Recomma stays with the answers — mostly captured from the interface a reader sees, each figure with its sample size and error bar, each fix re-measured fourteen days after it ships — and costs less to start. Teams that want to change what AI crawlers receive tend to fit Scrunch; teams that need to prove which changes moved the number tend to fit Recomma.

At a glance

Scrunch's figures are from its pricing page, month-to-month, in US dollars. Recomma's are from the plan table the product enforces.

Plans and prices

Scrunch sells two self-serve plans and Enterprise. Starter is $300 a month month-to-month, or $250 billed yearly, for 350 custom prompts, 1,000 industry prompts, three personas, five page audits and three users. Growth is $500 ($417 yearly) for 700 custom prompts, 2,500 industry prompts, five personas, ten page audits, five users and strategy reviews. Enterprise is priced on request and adds SAML or OIDC sign-in, the Enterprise Data API and a dedicated team. Extra seats are $25 a month each, and a team that nears its limits is contacted rather than blocked. Every plan includes agent traffic monitoring. Scrunch pricing.

Scrunch's own pricing FAQ describes a different line-up — a $250 Core plan with 125 prompts and four models, $500 Agency Core, and a trial that needs a card. We have used the pricing page, and suggest confirming the plan in writing before you buy.

Recomma sells three self-serve plans and an Enterprise tier quoted by a person. Starter is $30 a month for 50 prompts, sampled weekly, on one project. Growth is $299 for 300 prompts, sampled daily, across three projects. Scale is $999 for 500 prompts, daily, with no limit on projects or engines. Prompts pool across the workspace rather than being allotted per project. The full table is on the pricing page.

The two count prompts differently, so compare with care. Scrunch's 350 prompts on Starter are each re-asked every three days once they settle; Recomma's are re-asked daily on every plan. Scrunch also counts each platform's answer as its own data point, so a prompt tracked on seven engines yields seven answers per run. The plain difference is the entry price: Recomma starts lower, and Scrunch's cheapest plan buys more engines and users.

Engines, and how answers are collected

Scrunch's pricing page lists ChatGPT, Claude, Gemini, Perplexity, Google AI Mode, AI Overviews and Meta AI on every plan. Its platforms FAQ adds Microsoft Copilot and says Grok is coming soon; its data collection FAQ already counts Grok, for nine. That page says answers are gathered by "a combination of browser automation and official platform APIs", the method "selected per platform", without saying which platform uses which. New prompts are collected daily for fourteen days, then every 72 hours. Prompts can be tracked in any country and language, with geolocated results on ChatGPT, AI Overviews and Perplexity (international FAQ).

Recomma captures ChatGPT, Gemini, AI Overviews and Google AI Mode from the interface a reader sees rather than from a model API, and asks Perplexity through its own API. The difference is not cosmetic: in our study of ChatGPT, only 8% of the pages the interface cited also appeared in the API's answer to the same question (the interface vs the API). The cost of that choice is coverage: Copilot is not sampled, and Claude and Meta AI are not sampled at all. Each prompt is asked from one of 17 countries.

Agent traffic and AXP

This is where Scrunch is most different. Agent Traffic reads bot requests through integrations with Akamai, AWS CloudFront, Cloudflare, Fastly, Vercel and WordPress, and sorts AI visitors into training, indexing and retrieval for live questions. A Google Analytics 4 connection attributes human visits and conversions to the AI platform that sent them (agent traffic FAQ).

The Agent Experience Platform goes a step further. The CDN routes AI agents such as ChatGPT-User and PerplexityBot to AXP, which serves server-rendered HTML without JavaScript, with structured summaries and simplified components; people, Googlebot and Bingbot still get the normal site, and no code changes are needed (AXP FAQ). Scrunch says this can cut a page's token payload by up to 99% (AXP page). Neither page says which plans include AXP or what it costs.

Recomma takes the other route: it audits the real pages against fifteen rules that AI crawlers trip on and shows each failing page's markup as it is and as it should be, so the fix lands in the site everyone reads (the checklist). It does not monitor bot traffic. If a JavaScript-heavy site that cannot be changed quickly is the problem, AXP addresses it and Recomma does not.

What happens after the number

Scrunch reports brand presence, position, sentiment (positive, mixed or negative), share of voice and the sources answers cite, filterable by persona, funnel stage, region and topic. Share of voice is usually aggregated over 90 days to smooth daily swings; the page gives no sample sizes or confidence levels (share of voice FAQ). It also estimates AI search volume by topic from third-party panel data, which it calls "directional, not exact" (volume FAQ).

Recomma's additions are about whether a number can be trusted and acted on:

• Every figure carries the number of answers behind it and a 95% Wilson interval, and a reading too thin to mean anything is dimmed rather than drawn as a trend. Why the error bar.
• Each tracked question is priced with keyword data — monthly searches, the share of clicks a top result takes, and what advertisers bid — so you can see what the market behind a gap is worth.
• A gap becomes an action. When the action closes, the reading on its prompts is frozen, and fourteen days later the same prompts are asked again on fresh answers. The verdict is kept whether or not anything moved.
• Visits that assistants send to the site are reported beside the answers that sent them.

API, MCP and exports

Scrunch offers a Query API of aggregated metrics for BI tools and a Responses API of raw answers with their citations, each with 90 days of history, plus a Looker Studio template (API FAQ); the pricing page puts the Enterprise Data API on Enterprise. Its MCP server works on any plan, signs in through Scrunch's own login, reads metrics and answers, and can create and update brands, competitors, personas and prompts (MCP docs).

Recomma's read-only REST API comes with Scale; CSV export comes with every plan. Its MCP server, also on every plan, reads and writes: an assistant can list the gaps, read the site findings with their before-and-after markup, propose prompts and close an action. Every write previews first and needs a separate write scope, and new prompts land as suggestions that spend nothing until a person accepts them. Fixing what AI search gets wrong, from Claude Code walks through it.

Where Scrunch is the better choice

• Engine breadth. Seven engines on every plan, Claude and Meta AI among them. Recomma tracks five, and up to three below Scale.
• Agent traffic. Which AI bots visit, why, and from which CDN — Recomma does not report this.
• AXP. A served AI version of the site for teams that cannot quickly change the pages themselves.
• Markets and segments. Any country and language, with personas and funnel stages on every plan; Recomma offers 17 countries.
• Security paperwork. SOC 2 Type II, GDPR and CCPA (security FAQ).
• Track record and trial. Named enterprise customers, and a seven-day trial without a card. Recomma is new, has no reviews yet, and its trial needs a card.

Where Recomma is the better choice

• Price at the bottom. $30 a month to start, against $300 — daily, with up to three of its five engines.
• Answers as readers see them. ChatGPT, Gemini, AI Overviews and Google AI Mode captured from the interface, not an API, so the cited pages are the ones a buyer sees.
• Numbers you can defend. A sample size and confidence interval on every figure, so a change that is noise is shown as noise.
• Proof that a fix worked. Actions re-measured fourteen days after they close, against a frozen baseline.
• What the gaps are worth. Every prompt shows the priced demand behind it, from keyword data rather than topic volume from a panel.
• Fixes in the real site. A fifteen-rule audit with before-and-after markup, instead of a separate version for bots.

Choose Scrunch if… choose Recomma if…

Choose Scrunch if you need Claude, Copilot or Meta AI; you want to see which AI agents crawl the site and serve them a lighter version of it; you track many countries or personas; or your buyers ask for SOC 2.

Choose Recomma if ChatGPT, Gemini, AI Overviews and Google AI Mode and Perplexity cover the engines your buyers use; you have to show a client or a board which changes moved the number, with the error bar to back it; you want to see what the market behind each gap is worth; or your budget starts below $300.

For similar comparisons, see Recomma vs Profound and Recomma vs Peec AI; for the wider field, the best AI visibility tools.

How we compared

Scrunch's plans, prices, engines, collection method and integrations were read from its pricing page, FAQs and developer documentation on 14 October 2026, month-to-month in US dollars. Where its pages disagree — the plan line-up, the trial terms, the number of engines — we say so and lean on the pricing page. Where Scrunch does not publish something — which platforms use APIs, which plans include AXP, sample sizes — the table says so rather than guessing. Third-party review scores were left out because they could not be checked at the source. Recomma's figures are drawn from the same plan table its API enforces.