Write a competitive research plan before you collect the first fact

Public templates cover part of this. Aakash Gupta's competitive analysis template and RivalBeam's competitive intelligence checklist for PMMs both list what to track and how often to refresh it. Neither says how to record where a fact came from or when a fact is too old to use (both read September 14, 2026). Those two rules are what let you answer when a seller or an executive asks who said it and when.

Name the decision and its owner

Start with the decision. "Should we lead with price against Acme in mid-market deals" gives the research a stopping point. "Keep an eye on Acme" does not. Write the decision as one sentence, add the person who will make it, and add the date they need it by.

If the request arrives vague, ask the requester what will change depending on the answer, which segment or market it concerns, and what deliverable they expect. Claude can help here: paste the vague ask, have it propose two or three candidate decisions with the evidence each would need, then settle on one with the requester.

Reference: a vague ask and a scoped one (illustrative, with a placeholder competitor)

List the companies in scope and stop there

Name your own product and only the competitors the decision involves. Write a short stop-list as well: adjacent vendors, partners and analyst-only names that will surface in searches and that you will ignore for this plan. The competitor identification lesson covers how to choose the set when it is unclear.

The entity list also helps you manage usage. Claude's Research feature retrieves many sources per session and can use up your conversation limits faster than a normal chat, according to Anthropic's Research guide, checked September 14, 2026. If a run approaches your actual usage or time budget, narrow the company set again rather than turning it into an open market scan.

Rank the sources you will accept

List the source types you accept, then put them in order of trust for this decision. A useful default has four families:

1. The competitor's own published material: pricing pages, documentation, release notes, job postings.
2. Third-party material: review sites, news, analyst notes, community threads.
3. Your internal resources: past research, enablement documents, support tickets.
4. Field evidence: what sellers and customers reported in calls and lost-deal notes.

The order depends on the claim. For "what does Acme charge," the competitor's own pricing page outranks a seller's memory of a discount. For "why did we lose to Acme," field evidence outranks the competitor's website. Write the order per claim type if they differ, and write down what a source must include to count: a URL or document, a visible date, and enough context to see which product and plan it describes. A search snippet is a lead; open the page before it counts. A vendor-selected testimonial counts as the vendor's claim.

Set a freshness cutoff per claim type

Write the freshness rule down instead of judging each source by feel. Pricing and packaging change faster than leadership or funding, so set a separate cutoff for each claim type and treat anything older as unconfirmed until rechecked. Illustrative starting points, to adjust for your market:

• Pricing, packaging, discounts: 3 months
• Product capabilities and integrations: 6 months
• Positioning and messaging: 6 months
• Funding, leadership, headcount: 12 months

A fact past its cutoff stays in the file marked "recheck", with its original date, so the next session knows what to verify first. The conflicting research lesson covers what to do when a fresh source and an old one disagree.

Write the provenance convention

Decide the fields every collected fact will carry before anyone collects one. The minimum set:

• Source: the URL or document name, specific enough to reopen.
• Source family: competitor's own content, third party, internal, or field evidence.
• Date on the source, and the date you checked it. Both, because they differ.
• Scope: which product, plan, region and segment the fact describes.
• Source language, and whether a translation has been checked by a native reader.
• Who collected it, and the status: confirmed, unconfirmed, recheck, or retired.

The source-language field matters when Japanese pages are in scope. A pricing table read through Claude's translation is a lead until a native reader confirms the plan names, the tax treatment and the seat definitions. Keep the original passage beside the translation and have that check done before the claim reaches a battlecard. This is our recommendation for any cross-language claim; the Japanese competitor sources lesson covers where to look.

With the convention written, each fact becomes a row. The claim ledger lesson shows how to keep those rows and how to retire one.

Run the research inside the plan

Do the collection in a Claude Project set up as the project-setup lesson describes, with the plan saved as a knowledge file. Then turn on Research for the collection pass. Per Anthropic's Research guide, checked September 14, 2026, Research is available on paid plans (Pro, Max, Team and Enterprise), requires web search to be on, and starts from the "+" menu in a chat. On a free plan, run ordinary web searches yourself with the same scope and paste what you find into the chat.

Send Research a scoped request. One reusable form:

Prompt: a scoped research request

Use research-plan.md in this Project.
Decision: [one sentence]. Companies in scope: [names]. Ignore: [stop-list].
Collect only claims that support or contradict this decision.
For each claim, return: source URL, source family, date on the source,
product and plan it describes, and the exact passage.
Flag any claim older than the freshness cutoff for its type.
Report which claims you could not source instead of filling the gap.

Then open every citation it returns before using the claim. Check that the page still says what the passage says, that the date is visible, and that the product and plan match your scope. A claim without a working citation stays out of the deliverable; the verification lesson covers that check in detail.

Save the plan, then reuse it

Save the plan as research-plan.md in the Project's knowledge, next to the competitor files, with an owner and a checked date. At the start of every later session, restate the rule in the first message so Claude applies it instead of assuming it.

Revisit the plan when the decision changes, when a competitor enters or leaves the set, or when a freshness cutoff proved wrong in practice. Change the file, then replace the uploaded copy in every Project that uses it.

When the plan needs enforcing without you

The method above depends on one person restating the rules at the start of each session and rechecking dates by hand. It holds only while that person can keep up with the manual checks. It slips when several people research the same market, when battlecards are regenerated weekly, or when nobody remembers which pricing fact passed its cutoff last month.

In Steve's project settings, a team can set Source Priorities for Insights replies. Each project ranks four source types (internal resources, competitor content, on-field CI, third-party sources) in an order the team chooses.

Steve does not choose the decision or the competitor set; that scoping stays with you, and the plan above is still how to do it. If you want to compare Steve's source ranking with your current setup, start with Steve's competitive-intelligence product and confirm availability for your environment.