Profound: measure AI visibility before giving marketing agents authority

External-source research synthesis, not an OfferSignal client result. Based on the retained October 2026 research and source-specific dates.

A manufacturer can publish accurate capability information and still be absent, misdescribed or linked to the wrong evidence when a buyer asks an AI answer engine for suppliers. The first problem is discoverability and factual accuracy, not autonomous quoting. Profound began by measuring what answer engines say and expanded into workflows that help teams act on those observations while retaining human approval.

From answer measurement to governed action

Profound, founded by James Cadwallader and Dylan Babbs in New York in 2024, started as analytics for tracking prompts, answers, cited sources, competitors and visibility. Its later workflow adds Ask Profound queries, citations and sentiment, then lets marketers build no-code Agents. A workflow can scrape the web, call a large language model, query Answer Engine Insights, use APIs and retrieve knowledge-base material. Templates cover content refresh, frequently asked questions, competitive research and new content. Profound says Agents analyze cited pages and 16 reasoning models, connect to content-management systems and process hundreds of inputs through Sheets before human approval. 1 3 4 5

That is read/write marketing automation, not autonomous commercial authority. A person still defines prompts and goals, supplies product context, selects sources, reviews claims and approves publication. Background Agents, Profound Docs and additional media are described as forthcoming. The evidence does not show engineering validation, live capacity checks, configure-price-quote functionality or binding quotation. 4

What the named customer results actually show

Profound’s public cases contain specific visibility movements, but they are provider-published before-and-after observations. Ramp reports that two Accounts Payable pages generated more than 300 citations in one month and that measured AI visibility rose from 3.2% to 22.2% between 1 December 2024 and 15 February 2025, with rank moving from 19th to 8th among fintech brands. The case does not publish prompt count, repeated-run variance, confidence interval, traffic, pipeline, revenue or a control. An independent critique flags the same methodological gaps. 8 9 12

Statsig reports visibility increasing from 24.2% to 46.5% between 22 June and 20 July 2025 after comparison/category content and cited Reddit discussions. 1840 & Co. reports movement from 0% to 6% in two weeks and 11% by month-end after one targeted post, becoming a top-five cited remote-staffing brand. Neither case provides independent replication, repeated-prompt methodology or a qualified-enquiry denominator. These figures are visibility signals, not causal adoption, sales or return on investment. 10 11

An illustrative industrial application

This is an illustrative application, not a reported Profound deployment. An electrical-component producer could create a governed prompt set covering capabilities, materials, applications, comparisons and request-for-quote questions. For each run, record whether the company is mentioned, which URL is cited, whether the description is correct and whether the answer uses the current revision. Source data should expose canonical identities, equipment, materials, tolerance ranges, capacity windows, project evidence and quote boundaries.

The technical or commercial owner remains authoritative. Profound can identify missing evidence, draft a capability page or route approved content to a content-management system and work tools. It should not infer live capacity, approve substitutions, validate a drawing, price a custom job or issue an offer. Test the same prompt set repeatedly, split by buyer intent, and connect visibility to human-qualified enquiries rather than treating a citation as a sale. 3 13 14

Cost, controls and the decision boundary

The public offer includes a free seven-day trial with 50 prompts per day across ChatGPT, Gemini and Google AI Overviews. Enterprise pricing is custom; public pages do not state credit-unit, setup, implementation, integration, migration or overage fees. Budget for review, content-management work and repeated measurement. 2

Use least-privilege connectors, role-based access, approval before publishing, versioned prompt and source logs, source allowlists, duplicate-action checks, malicious-retrieval filtering, and expiry for price and capacity claims. Before scaling, request repeated-run methodology, referral and enquiry denominators, export and application-programming-interface terms, retention and data-residency terms. Profound is best treated as an AI-discovery and accuracy-testing layer: valuable upstream of an industrial enquiry, but not a replacement for the owner who makes the consequential decision.

Sources and next reading

Profound homepage · Pricing and trial · Profound Agents · Integrations · Vercel Agent Analytics

Explore OfferSignal · Focused Business & AI Workshop · All research

References