Profound

Profound is not simply an AI-visibility dashboard. It combines a daily monitoring panel, a licensed dataset of prompts submitted by real users, factual-claim checking against a controlled Knowledge Base, traffic attribution and workflow agents that can prepare or execute remediation. That breadth is its commercial advantage—and the reason a buyer must separate four different meters before comparing its price.

> Distinctive strength: Profound connects real-prompt demand research and citation-level factual-error diagnosis to executable marketing workflows in one governed platform. > > Where it stops being an advantage: Starter tracks ChatGPT only; Growth is limited to three engines and three seats; API and Shopping require Enterprise; Agent-credit overage pricing is not public.

The buying question is: does your organization need the complete research-to-action loop, or would a narrower monitor produce the same decisions for less? Profound's panel model, FactCheck accuracy and citation impact still need to be demonstrated on the buyer's own corpus. This page defines that proof rather than treating the breadth of the platform as proof by itself.

Who can justify Profound's operating model?

What are you actually buying?

Profound spans four operating layers:

1. Monitor: Answer Engine Insights runs a configured prompt panel daily and extracts mentions, position, citations, sentiment and competitors.
2. Research: Prompt Volumes analyzes a licensed panel of real-user prompts to identify demand and missing questions.
3. Verify: FactCheck extracts atomic claims, compares them with a customer Knowledge Base and links inaccuracies to cited sources.
4. Act: Agents and integrations can produce content, reports, outreach or external API/CMS actions.

Buying only the first layer makes Profound comparable to Otterly, Peec or Rankscale. Buying all four makes it an operating platform, with more value potential and more governance work.

Four meters that must never be combined

Any cost comparison that divides the subscription by one meter and applies the result to another is misleading.

How much does Profound cost?

Both public prices are tied to annual billing and advertise two months free. Growth adds CSV/JSON exports but no API. Enterprise gates API access, SSO/SAML, multiple companies, Shopping, custom regional/language coverage and broader support.

What the response figures mean

Starter's 50 prompts × one engine × 30 days yields 1,500 response observations. Growth's 100 prompts × three engines × 30 days yields 9,000. These are arithmetically transparent, but they remain observations from a controlled panel—not a count of what consumers saw.

The nominal subscription cost per response observation is about 6.6 cents on Starter and 4.4 cents on Growth. This is only a capacity normalization. It says nothing about model accuracy, decision quality, Prompt Volumes, FactCheck or Agents.

Are Prompt Volumes real search demand?

Profound says Prompt Volumes is built from licensed, double-opt-in consumer panels containing real prompts from ChatGPT, Gemini, Claude and Perplexity. It says synthetic prompts are not used. The vendor aggregates, anonymizes and cleans the data, then applies probabilistic modeling for demographic and geographic biases.

This is materially different from a vendor generating its own prompt set and repeatedly submitting it to models. It is still not a census.

Important limitations disclosed by Profound:

• data updates weekly with less than two weeks of latency;
• coverage is strongest in the US, UK and select European markets;
• US/ChatGPT history begins January 2025; other regions/platforms generally July 2025;
• regional engine coverage varies;
• early periods can be ChatGPT-heavy;
• demographics are modeled from panel data.

The correct description is modeled demand from an opt-in panel, not “all AI searches.” A buyer should compare panel topics with sales calls, site search, support tickets and first-party research before prioritizing content.

What does FactCheck actually verify?

FactCheck uses prompts tagged for factual review, extracts atomic claims about the brand and compares them with a customer-maintained Knowledge Base. It classifies each claim as accurate, inaccurate or not relevant and shows associated citation URLs. Subjective claims are routed to sentiment instead.

This workflow can be valuable for pricing, terms, product capabilities and regulated facts. It has one non-negotiable dependency: the Knowledge Base must be more reliable than the answers being checked. Outdated or contradictory source material turns automation into a false-confidence machine.

The product allows users to correct false classifications. That is good product behavior and evidence that human review remains part of the system, not an optional inconvenience.

A reproducible FactCheck scorecard

Build a test Knowledge Base with 40 current atomic facts and ten deliberate outdated conflicts. Create prompts likely to elicit each fact. Manually label every resulting claim and calculate:

• precision: how many “inaccurate” flags are truly inaccurate;
• recall: how many known inaccuracies are found;
• citation traceability: how often the blamed URL plausibly supports the false claim;
• remediation usefulness: how many findings map to a specific owned or third-party action.

Do not buy FactCheck on the basis of an accuracy score produced by the same system being evaluated.

How do Profound Agent credits work?

Starter includes 100 Agent credits monthly; Growth 400. Each run consumes credits according to workflow complexity. Profound shows an estimate before running and the actual debit afterwards. Organization settings expose runs, average credits, users and consumption by Agent/Sheet.

This is better than invisible compute charging, but the public evidence does not provide a universal “one credit equals X nodes/tokens” formula. Complex multi-step Agents consume more. Accounts can either pause at the allocation or continue with overage billing. A public overage unit price was not established.

The safe cost model is:

monthly platform cost + Agent overage + human review + integration maintenance + remediation delivery

A successful Agent run is not a successful editorial action. Measure the accepted and published output after review.

How should Agent workflows be governed?

Profound Agents can combine product data, LLM steps, Sheets, native integrations and external API calls. A Call API node supports GET/POST and several authentication methods. This flexibility means the blast radius can extend beyond a dashboard.

Use four stages:

1. read-only data collection;
2. draft-only recommendation;
3. named human approval;
4. restricted write/action with audit log and rollback.

Never let a sentiment change or FactCheck label automatically publish content, contact a third party or update a CMS. A classification error must stop at the approval boundary.

What is Enterprise really buying?

Enterprise is not just “more prompts.” It is the route to up to nine engines, custom language/region coverage, multiple companies, API, Shopping, SSO/SAML, a SOC 2 claim and dedicated Slack/specialist support.

The pricing matrix lists a dedicated specialist with a 24-hour SLA. That appears to be a support response commitment, not a public uptime SLA. Obtain the exact SOC 2 report scope, service availability terms, subprocessors, data flows and Agent overage schedule during procurement.

Privacy and data boundaries

Profound's public privacy policy names Cooper Square Technologies Inc. as operator. It describes the company as controller for website/account activity, while customer-platform data processed on behalf of customers is governed by customer contracts.

The policy permits disclosure to service providers, including AI service providers. It also allows creation/use of anonymized or aggregate data and lists broad retention exceptions. A serious buyer should obtain the DPA and subprocessor list and map which prompts, Knowledge Base records, Agent inputs and connected-system data reach which providers.

The public policy does not establish one fixed deletion clock for every product record and backup. Procurement must resolve this for confidential or regulated content.

When should you exclude Profound?

Exclude or postpone it when:

• Starter's ChatGPT-only coverage cannot represent the audience;
• Growth's three engines or three seats are insufficient but Enterprise cannot be justified;
• API, Shopping or SSO is required without custom procurement;
• the team cannot maintain a current Knowledge Base;
• no human approval can sit before Agent writes/calls;
• Agent overage price is needed for budgeting but remains undisclosed;
• panel demand would be presented as complete global search demand;
• the organization only needs a compact visibility baseline.

Test the workflow, not the dashboard demo

1. Freeze 30 commercial prompts and the expected brands, facts and citations.
2. Run the monitoring panel for seven days; manually audit 100 responses.
3. Compare Prompt Volumes topics with the panel and with first-party demand signals.
4. Record where panel coverage differs by region or engine.
5. Build a Knowledge Base with 40 current facts and ten controlled conflicts.
6. Calculate FactCheck precision, recall and citation traceability.
7. Create one read-only Agent and record estimated vs. actual credits across ten runs.
8. Add a draft-only output step; score factual corrections and editing time.
9. Place a human approval gate before any API/CMS/communication action.
10. Test exports, roles, offboarding and evidence custody.
11. Review DPA, AI providers, SOC 2 scope, retention and support/uptime terms.
12. Calculate total cost separately for monitoring, prompt research, Agent compute and labor.

Pass only when at least three findings lead to accepted, auditable actions and the complete process remains inside the approved data/cost boundary.

Profound or a narrower visibility monitor?

The proof required before a Growth commitment

Do not begin with an Agent. First prove that the monitoring and Prompt Volumes layers reveal a valuable gap. Then test FactCheck against controlled facts. Only after those two gates pass should the team spend Agent credits or connect a write-capable system.

Continue through the AI SEO software directory, compare fixed-panel economics in Peec AI, and use the forthcoming guide to validating AI visibility data before attributing an outcome to the platform.

Our Profound verdict

Profound has the broadest evidence-to-action story in this first cohort: real-prompt panel research, daily visibility monitoring, claim verification, citations and executable workflows. That can justify the price for a team mature enough to govern the full loop.

For a small team that only needs recurring brand/citation observations, it can be unnecessary complexity. The honest conversion path is not “buy Enterprise.” It is: prove monitoring value, validate the panel, test FactCheck, then earn the right to automate.

Official sources

• Profound pricing and plan comparison ↗
• Prompt Volumes methodology ↗
• Prompt Volumes FAQ ↗
• Creating monitored prompts ↗
• FactCheck ↗
• FactCheck practices ↗
• Agent credits ↗
• Integrations and connectors ↗
• Call API node ↗
• Agency Mode ↗
• Shopping ↗
• Profound privacy policy ↗
• Profound service status ↗

Sources checked 2026-08-28. Pricing, coverage, panel methodology, credits and terms can change; recheck them before publication or procurement.