Comparison · Evidence checked 2026-08-28

Profound vs Scrunch AI: Prompt Intelligence, Agents or AI-Facing Delivery?

A practical, evidence-led decision guide. Product capabilities and limits are separated from anything that would require hands-on testing.

Profound and Scrunch AI are not simply expensive versions of a prompt tracker. Profound builds toward an intelligence and action system: monitored answers, licensed prompt-demand data, claim checking against a Knowledge Base and governed agents. Scrunch combines monitoring and page audits with agent-traffic analysis and AXP, a delivery layer designed to serve approved content to AI agents.

The products therefore create different organizational commitments. Profound asks a team to govern research, facts and agent actions. Scrunch can ask engineering and brand owners to govern an additional AI-facing representation of content. Buy neither until those owners exist.

Choose by the operating model

Requirement	Profound	Scrunch AI
Find which prompts may matter beyond a manually curated list	Stronger fit through Prompt Volumes	Industry prompts can broaden coverage, but retained evidence does not establish the same demand-data method
Monitor answers and citations	Yes, with plan-specific engine and response limits	Yes, through custom and industry prompt surfaces
Check specific claims against approved facts	FactCheck plus a controlled Knowledge Base	Page audits and recommendations are the more visible starting point
Automate research or remediation	Profound Agents, with separate credits	Workflow depends on Scrunch surfaces and enterprise implementation
Observe AI-agent traffic	Not the central retained proposition	A material Scrunch use case
Serve a controlled AI-facing content layer	Not the central product path	AXP is the distinctive Scrunch architecture
Self-serve pricing that can be modeled today	Clearer, though annual billing and separate meters matter	Blocked by conflicting current official plan descriptions until checkout is reconciled

Profound is the more coherent choice for a research-to-action programme. Scrunch is the more differentiated choice when the buyer wants monitoring connected to how AI agents access the site. That second path carries the larger technical and brand-governance burden.

Profound has four separate economies

Profound’s public self-serve tiers should never be summarized as “$99 for 50 prompts” or “$399 for 100 prompts.” Four meters govern different jobs:

Monitored prompts: 50 on Starter, 100 on Growth.
Response observations: 1,500 and 9,000 per month.
Prompt Volumes: a separate licensed demand dataset, absent or limited according to tier.
Agent credits: 100 and 400, consumed according to workflow complexity.

Starter is listed at $99 per month billed yearly, with ChatGPT, one seat, 50 prompts, 1,500 responses and 100 Agent credits. Growth is $399 per month billed yearly, with three engines, three seats, 100 prompts, 9,000 responses, 400 credits and CSV/JSON export. API, Shopping, SSO/SAML, multiple companies and broader engine coverage are Enterprise gates.

An organization can have enough response capacity and too few Agent credits. It can also have enough monitored prompts but no Prompt Volumes entitlement for discovering demand. Model each pool separately.

Scrunch has a pricing conflict before it has a cost model

Scrunch’s current canonical pricing page lists Starter at $300 monthly or $250 on annual terms, with 350 custom prompts, 1,000 industry prompts, three personas, five audits and three users. Growth is $500 monthly or $417 annually, with 700 custom prompts, 2,500 industry prompts, five personas, ten audits and five users.

A separate official FAQ still describes Core at $250 with 125 prompts and four LLMs, plus an Explorer trial. These are not small wording differences. They describe different plan structures and cannot both serve as the basis for a current invoice.

Treat Scrunch pricing as conflicted until the buyer captures the live checkout or obtains a dated written quote covering custom prompts, industry prompts, personas, audits, users, engines, AXP, API, SSO, renewal and trial terms. Do not average the tables or select the newer-looking one by intuition.

The number that changes the Profound decision

Profound Growth includes 9,000 monthly response observations for 100 prompts across three engines. At daily cadence, the visible dimensions multiply to:

`100 prompts × 3 engines × 30 days = 9,000 responses`

That is internally coherent with the stated response allowance. Starter’s 50 prompts against one engine daily similarly imply 1,500 responses.

The agent economy is unrelated. If a ten-run workflow consumes an average of six credits per run, it uses 60 credits. Repeat it weekly and the monthly requirement is about 240 credits, which exceeds Starter’s 100 even though monitoring still fits. The six-credit figure must come from an observed test; public evidence does not justify inventing a universal cost per run.

The number that changes the Scrunch decision

Scrunch separates custom prompts from industry prompts and also adds personas. Before procurement, establish whether a persona causes an additional scheduled answer, a reporting segmentation only, or another kind of consumption. Then calculate custom and industry coverage independently.

A team might need only 80 carefully chosen commercial prompts. In that case 350 custom prompts are not automatically valuable, while five audits may be the scarce allowance. A larger industry-prompt pool is useful only if its composition, market relevance and stability can be inspected.

The practical capacity is the first exhausted meter among custom prompts, industry prompts, personas, audits, users and any AXP or enterprise entitlement.

Profound’s distinctive path: demand to governed action

Profound says Prompt Volumes uses licensed, double-opt-in panel data that is anonymized and aggregated, with no synthetic prompts. The data is modeled for demographic and geographic bias, freshest weekly with less than two weeks of latency, and strongest in the US, UK and selected European markets.

This can improve a manually invented monitoring set. It does not establish a complete global count of AI-search demand. Test whether the panel identifies commercially relevant questions missing from the team’s own list and record where market coverage is weak.

FactCheck adds a second differentiated surface. A buyer can build a Knowledge Base of atomic facts and test whether monitored answers contradict them. The result is only as good as the approved fact set and classification. Outdated truth in the Knowledge Base can make a correct answer look wrong.

Agents then create a path from finding to research or remediation. This is where governance matters most. Start read-only. Require human approval before a CMS write or external API call. Preserve the observation, source, Knowledge Base fact, Agent inputs, generated output, credit consumption and approval decision.

Scrunch’s distinctive path: evidence to AI-facing delivery

Scrunch brings together response monitoring, page audits, agent-traffic evidence and AXP. The useful idea is that the team can diagnose what answer systems see and then provide a controlled, current representation intended for machine consumption.

AXP should not be evaluated as a marketing toggle. It is infrastructure. The organization needs an approved source of truth, ownership rules, freshness controls, parity checks against canonical content, change review, observability and rollback. An AI-facing layer that drifts from the public page can create two competing versions of the brand’s facts.

The first Scrunch trial should therefore stop before AXP activation. Prove the monitoring, citations, audits and agent-traffic attribution. Only then run a separate architecture review using non-production content and explicit rollback criteria.

One procurement test, two different proof burdens

Freeze 30 commercial prompts across discovery, comparison, pricing, risk and alternatives. Choose the same three engines where possible and retain raw answers, cited URLs, model/surface, locale and timestamp.

Profound proof pack

manually audit 100 answer classifications and citations;
compare Prompt Volumes topics with the frozen panel and label useful additions;
create a 40-fact Knowledge Base with ten deliberate outdated conflicts;
calculate FactCheck precision and misses against human labels;
run one read-only Agent ten times and record actual credits and output quality;
verify that no write or API action can bypass approval;
export and reconcile stable identifiers.

Scrunch proof pack

reconcile custom and industry prompts with the documented plan;
establish how personas affect measurement and consumption;
manually audit response and citation extraction;
compare five page audits with an independent technical/editorial review;
inspect agent-traffic attribution using controlled synthetic identifiers;
document every recommendation’s source evidence;
map AXP architecture and rollback without enabling it in production.

Compare the number of verified findings, accepted actions, analyst hours and unresolved governance gates. Do not compare proprietary visibility scores.

Security and procurement boundaries

Profound’s Starter and Growth tiers gate several enterprise controls. Teams needing API, SSO/SAML, multiple companies or broader engines should price Enterprise from the start. Review the DPA, subprocessors, contracted AI providers, SOC 2 scope, retention, deletion and support commitments. A public status page is useful operational evidence, not a contractual SLA.

For Scrunch, obtain the current security pack, retention schedule, API/SSO entitlements and exact data flows for monitoring, agent traffic and AXP. Determine what content is copied, transformed or cached, where it is served, how quickly canonical corrections propagate and what survives account cancellation.

Neither vendor’s customer outcomes prove your programme will create citations. Treat case studies as hypotheses and retain your own before/after panel.

When to exclude Profound

Exclude Profound when Starter’s single engine or Growth’s three engines cannot cover the programme, Enterprise-only controls are mandatory but unaffordable, Prompt Volumes geography does not match the market, or nobody can maintain the Knowledge Base and review Agent work. Also pause when Agent overage economics are required but unavailable.

A team that wants a simple weekly mention report is likely to pay for capabilities it will not govern.

When to exclude Scrunch

Exclude Scrunch when official checkout cannot reconcile the plan conflict, when custom and industry prompt meters remain unclear, or when AXP would create an unreviewed alternate truth. Also stop if agent-traffic attribution cannot be reproduced or the organization lacks technical ownership for delivery changes.

A team that only needs monitoring should not adopt an AI-facing delivery layer to justify the platform price.

Our verdict

Profound is the stronger research-to-action system. Its four meters are complex but legible, and the chain from licensed prompt demand through monitored evidence, FactCheck and governed Agents is genuinely distinctive. It earns a shortlist when the organization can maintain the controls around it.

Scrunch is the stronger technical proposition when the programme includes page auditing, AI-agent traffic and controlled machine-facing delivery. That can be valuable for a large brand, but the unresolved pricing conflict and AXP governance burden are hard gates, not footnotes.

Choose Profound when the accountable owner is an intelligence, SEO or content-operations team. Choose Scrunch when engineering, brand and governance teams jointly own how content reaches AI agents. Choose a narrower monitor when neither operating model exists.

Use the AI visibility monitoring cost guide to separate their meters and the validation protocol to build the shared truth set.