Competitive Intelligence Report Prompts

Read any competitive report closely and the lines split into two kinds. Some describe something that happened: a price moved, a page changed, a deprecation was announced. The rest describe how things stand: this is how they position themselves, this is who they sell to, this is what their pricing looks like.

The first kind has a date built into it. The second kind has no date at all. There is no day on which a competitor’s positioning is true. What exists is a day somebody checked, and that day is usually missing from the report.

So a reader gets one page carrying a fact confirmed six weeks ago and a fact last confirmed seven months ago, in the same type, under adjacent headings. Both read as the current picture, because a report is presented as the current picture. Nothing on the page distinguishes them.

A standing claim needs the date it was last true

The fix is one field and it is unusual enough to be worth stating plainly. Every standing claim carries the date it was last verified, which is the date you captured the source it came from. Not the date of the report.

That single addition changes how the document gets read. “They position themselves as an agentic platform, last verified February” invites somebody to go and look again. The same sentence without the stamp invites nothing, because there is no visible reason to doubt it. A filing is the one source that dates itself, which is why SEC filings are the easiest standing claims to keep honest.

Every one of these documents has a section headed something like “what changed this period”. It is the section people read. And a model has no way to know what the period is unless the prompt says so.

Left unstated, “this period” resolves to everything you pasted. That sounds harmless until you look at a real pile of sources. Filings are annual. Earnings calls are quarterly. Changelogs are weekly. A folder of material collected for one monthly report routinely spans seven months, and every item in it is legitimate evidence.

The result is a September report in which a February statement appears under what changed, correctly sourced and correctly dated, next to something from three weeks ago. Nobody lied. The reader draws a conclusion about momentum that the material does not support.

Two dates in the prompt fix it, and they buy a second thing as well. Once the period is explicit, anything outside it can go into a block of its own: still true, still worth having, not news. That block is where standing claims belong, and it is the part of a report most teams have never separated out. Product is the dimension where the two blocks diverge fastest, because a competitor's roadmap produces dated events and a feature list produces a state.

Four inputs, and only the first is the pile of sources. The other three are decisions you make before the assembly starts, and each one takes a line.

The fourth is the least obvious and it does the most for the reader. Naming the sections you expect, including the ones you suspect are empty, is what makes an absence visible. A section nobody named does not come back empty. It does not come back at all.

Four prompts, and the order is not the order you would guess. The coverage check runs first, before anything is assembled, because what it finds usually sends you back to collect rather than on to write.

The third prompt uses last period’s report, and for a different purpose than you might expect. Not to avoid repeating the news, which is what a recurring competitor digest needs it for. Here it supplies the standing claims, so this period’s material can confirm them, contradict them, or say nothing about them.

Silence is the usual answer and it is the useful one. A claim nothing has contradicted for six months is not wrong. It is unverified for six months, which is a different thing and is the thing nobody writes down.

Standing claims are also easier to keep somewhere other than the report. Hold them in competitor profiles and the report can link to them rather than restate them, which is what stops a reader learning to skim the middle four sections.

We assembled a report from ten labelled items on one public competitor. Nine carried a date. One was undated on the source page, and was labelled as undated in the material. The items spanned February to September 2026, and the first run deliberately named no period.

The undated entry borrowed a date from its neighbour

Item ten was six product entries grouped under a season name, with no date on any of them. The report printed them with a date: 15 September, taken from the changelog announcement sitting next to them in the material.

The date is not absurd. It is the day that release was announced elsewhere, so somebody checking casually would find it plausible. That is the expensive part. A wrong date that looks wrong gets queried; a wrong date that looks right becomes the date, and it is the date a battlecard inherits three weeks later.

The honest positive from the same run is worth recording. Asked for pricing and for customer evidence, having been given neither, the report said so twice and padded nothing. The instruction to name what is missing worked exactly as written. The failures were both about dates, which is the one thing the format cannot police on its own.

“Nothing to report this period” is one of the most useful sentences a competitive report can carry, and one of the most dangerous, because it is written identically in four quite different situations.

A gap named is a gap somebody can close

The coverage check exists for the second row. Run before assembly, it lists the sections with no evidence behind them, which is the one moment in the process where that list is still actionable. After assembly the same gap has become a sentence in a document and stops looking like work anybody has to do.

The fourth row is the one people miss, and it is also the most interesting. A page that disappears, a metric a company stopped reporting, a comparison it took down: each is a decision somebody made, on a date. A report that only records what appeared will never carry any of them, which is why the useful competitive signals include absences.

The second row is the one to act on, and it usually names one dimension rather than several. Hiring is the common case: a month where nobody read a competitor’s open roles reports as a quiet month for hiring, and the two are indistinguishable on the page.

The expensive half of this job is not the writing. It is arriving at the start of the period with sources you actually captured, each with a date, across every dimension you promised to cover. Most reports are late and thin for the same reason: the collecting happened in the last three days.

That is also why standing claims go stale invisibly. Re-checking a competitor’s positioning when nothing prompted you to is the least satisfying work in the discipline. So it does not happen. The claim carries forward for another period with no visible age on it.

Some of this no tool touches. What a period means for your plan, which decisions it forces, and how much any of it matters are judgements somebody has to make. But intelligence freshness is worth measuring precisely because it degrades without anybody noticing.

Flares watches competitor pricing, product, documentation and messaging and stamps a date on every change it finds. A report then starts from a dated record rather than from a folder, and a claim carried forward from last period can be re-verified rather than retyped. Where the request came from another team in the first place, a request brief is what says which sections they actually needed.