How a research run works, step by step
An AI research agent for competitive intelligence does the job in three stages: profile, monitor, alert. The first run builds the profile. The agent reads a competitor's public pages, pulls pricing, positioning and recent moves into one structured document, and stores it. That run costs 3 credits, and the price shows at the approval gate before anything fires, so you know what a piece of intel costs before you buy it.
Monitoring is a lighter run. A sweep costs 1 credit and compares what is live today against the stored profile. You set the cadence in plain words: one sentence in chat becomes a scheduled run, and it fires with your laptop closed. Weekly suits pricing pages, which change quietly and matter immediately. Monthly is usually enough for positioning and messaging.
The alert is a diff, not a document. When a sweep finds a change, you get what changed, what it said before, and the source link so you can check it yourself in ten seconds. When nothing changed, you hear nothing, which is what keeps a monitoring schedule readable after week three.
How intel moves into outbound, content and ads
Intel lands in shared memory in one of two states. A hypothesis is something observed but untested: a competitor raised prices, so their cost-sensitive customers might be movable. A lesson is a hypothesis backed by a measured outcome: the price-change angle got replies, so it becomes the default opener. Every discipline reads the same memory, so nothing gets rediscovered twice and a stale angle gets retired everywhere at once.
That is the practical difference from standalone competitor monitoring. The same agent that spotted the price change can draft the outreach batch that uses it (1 credit for 5 drafts), write the comparison page (3 credits), or fold it into the next ad variants. Each of those runs stops at the approval gate. You read the drafts, then they go. The intel-to-action handoff that normally dies in a shared doc simply is not a handoff here.
What to monitor first
If you are starting from zero, four public pages carry most of the signal. The pricing page, because price moves force a response and competitors rarely announce them. The changelog or release notes, because shipping cadence tells you where their engineering attention is. The homepage headline, because a rewrite there means a repositioning is underway. And the careers page, because roles they open show what they plan to build next.
Start with your two or three closest competitors, not ten. Every sweep costs a credit, and the goal is a picture you actually read, not coverage you archive. Widen the list when an alert changes a decision you were about to make; that is the test that a competitor is worth the credits.
What it costs to keep the picture current
Most competitive intelligence platforms do not publish prices; you book a demo and get a quote. Here the math fits in one paragraph. A full competitor profile is 3 credits, once per competitor. A monitoring sweep is 1 credit, so a weekly schedule runs about 4 credits a month. Profile three competitors and keep weekly sweeps on and your first month lands well inside a Starter plan, with room left for the runs that act on what you find.
The credit rules keep it predictable. Conversation is free, only runs bill. Failed runs refund automatically. Unused credits roll over up to 2x your plan. You get usage alerts at 50, 80 and 100%, and overage bills at 1.25x with a hard cap at 3x, so a forgotten schedule cannot quietly triple your bill. The 14-day trial includes 25 credits, enough to profile a few competitors and see the first sweeps come back before a plan starts.
When a dedicated competitive intelligence tool is the better fit
Some teams should buy a dedicated platform, and this page will not pretend otherwise. Enterprise competitive intelligence tools pair software with human analysts who validate every signal, and they push battlecards straight into a large sales team's CRM and chat tools. If you run a sales org that lives on battlecards during deal cycles, that workflow is mature and worth paying for.
Moatt fits the other case: a small team where intel is only useful if it turns into outreach, pages and ads without leaving the system. You get published prices, a visible cost per run, and one memory shared across disciplines instead of a standalone research function producing reports for someone else to act on.