# When competitor sources disagree, rank them by authority, then by date

Decide what a battlecard may say when two competitor sources disagree or a verified claim has aged: normalize, rank, show both or withhold, set a recheck date.

URL: https://www.hiresteve.ai/blog/conflicting-research
Updated: 2026-10-12

When two competitor sources disagree, first confirm they describe the same plan, market, unit and period. If the conflict survives, rank the sources by how close each sits to the fact, let the newer one win only between equals, and write the disagreement into the battlecard with both dates. Withhold the claim when neither source is good enough, give every changeable fact a recheck date, and hand pricing, legal and reputational calls to a named owner.

This lesson starts where the [claim ledger](/blog/competitive-claim-ledger) and the [citation check](/blog/verify-ai-research) stop: each claim already has a row with a source, a date and a confidence. What remains is two sourced rows that disagree, or one verified row that has aged past the point where anyone should repeat it.

## Confirm the two claims describe the same thing

Some apparent contradictions are scope mismatches. [Digital Applied's guide to conflicting research sources](https://www.digitalapplied.com/blog/ai-research-conflicting-sources-guide) makes the point for research writing in general: a price-tier or launch-date mismatch may be a scope mismatch rather than a true conflict, and normalizing comes before judging. Put the two ledger rows side by side and compare four fields.

- **Scope.** Plan tier, region and edition. An enterprise plan and a self-serve plan can both be described truthfully with different numbers.
- **Period.** When each source was published, and which period it describes.
- **Unit.** Per seat or per workspace, monthly or annual, list price or negotiated.
- **Subject.** The same product under the same name. Rebrands and acquisitions produce many false conflicts.

An invented pair for practice: a pricing page lists a Pro plan at one per-seat monthly rate, and an eight-month-old analyst note lists a lower figure that turns out to be the annual-billing rate. That is one price with two billing terms, so both rows get corrected rather than ranked. Only a conflict that survives this step moves on.

## Rank by authority, then let the date decide changeable facts

For current product facts, start with distance from the fact. The competitor's own current pricing page, release notes and documentation sit closest. Dated independent evidence, such as a product review, a reseller listing or a captured screenshot, sits one step out. A forum comment, a seller's recollection or a summary by someone who never opened the original sits furthest.

Recency decides only between sources of similar authority. For facts that change often (prices, plan limits, feature availability, integrations) prefer the newer source. For slow-moving facts such as founding year or stated target market, an older primary source still beats a newer secondary one.

Watch for two traps. A newer source can be wrong because it copied an older one; three blogs quoting the same outdated figure count as one source. And a vendor's marketing page can lag its own documentation, so when the two disagree, "the vendor said so" does not end the check.

## Run the comparison in Claude, then open both sources yourself

Work inside the [Claude Project you set up for product marketing](/blog/claude-project-setup), so the competitor files and approved wording are already in context. Paste both ledger rows with the supporting passage from each source, and ask for the scope check before the ranking. This is the one instruction in the lesson whose exact wording changes the result.

```prompt
Here are two claims about the same competitor, each with its source and date.
First check whether they describe the same plan, market, unit and period.
If they do not, say what differs and stop.
If they do, say which source is more authoritative and which is more recent,
give your reasoning in two sentences, and recommend one of:
prefer A, prefer B, show both with dates, or withhold the claim.
```

Then open both cited pages. Claude's ranking is a suggestion about which page to trust, and it stays unchecked until you have read that page. Write the decision, the reasoning and the recheck date back into the ledger row so the next person does not repeat the work.

For a one-off pull of everything recent about a competitor, use Claude's Research mode. Anthropic's [Research help page](https://support.claude.com/en/articles/11088861-use-research-on-claude), read September 16, 2026, lists it for paid plans (Pro, Max, Team and Enterprise) on web, desktop and mobile with web search enabled, and describes its output as citations you can check. The companion [guidance on web search, extended thinking and Research](https://support.claude.com/en/articles/11095361-when-should-i-use-web-search-extended-thinking-and-research) describes Research comparing competitors and updating outdated internal documents.

Neither page promises automatic adjudication between two cited findings or automatic staleness detection. The adjudication is yours. Claude can present training-data recall about a competitor with more confidence than the evidence supports. A prior fetch is not a version history: have Claude fetch the current page and compare it with a saved prior snapshot or another dated source, then verify the changed passage yourself.

## Show the disagreement, or withhold the claim

After ranking, a battlecard has three honest states. It states the better-supported claim with its date and source. It shows both claims with both dates, when the sources are of similar authority and the difference matters to a seller. Or it leaves the claim out, when neither source is fit to put in front of a customer. The Digital Applied guide above reaches the same conclusion for research writing: disclose the conflict or withhold the claim.

**Averaging and quiet picking both fail**

Averaging two figures produces a number no source published, and a seller who repeats it cannot defend it. Quietly choosing one source and deleting the other hides a known disagreement from the person who will be challenged on it. If you can neither show both nor withhold, the claim is not ready.

A disclosure fits in one line: "Pro plan listed at [price] on the vendor's pricing page as of [date]; a [month, year] reseller listing shows [other price], unconfirmed." The losing source stays in the ledger with the reason it lost, so nobody reopens the argument next quarter.

## Put a recheck date on anything that ages

A claim can be sourced, uncontradicted and still unsafe, because the world moved and nobody looked. Staleness is a separate check from disagreement, and it runs on a calendar rather than on a complaint.

Give each changeable claim its own recheck date when you write the row. As an illustrative starting schedule, use a quarter for prices and plan limits, a month for feature availability while a competitor is shipping quickly, and a year for slow facts such as ownership or target market; adjust each interval to the change history and risk you observe. The Japanese battlecard guide at [compato.app](https://compato.app/blog/battlecard-creation-and-update-workflow) puts price facts at about three months and notes that a competitor shipping monthly or every two months can have a very different feature set within about six months; it argues that stale battlecard information can become counterproductive as it ages. It offers no measurement, so treat those intervals as one practitioner's view.

When a recheck date passes, reopen the source before deciding anything. If the page still says the same thing, update the checked date. If it has changed, the new page and the old row are now a contradiction, and the ranking above applies. If the page is gone, the claim drops to unverified until you find a replacement source.

## Escalate pricing, legal and reputational claims

Some claims should not be settled by one person's view of which source is better. Escalate to a named owner when the claim would appear in a customer-facing price comparison, when it says something negative about a competitor's security, compliance or legal standing, or when a seller has already told a prospect the old version. Record in the row who decides, by when, and what the card says meanwhile. "Under review, do not quote" is a legitimate battlecard state.

- Both claims describe the same plan, market, unit and period, or the mismatch is written down.
- The more authoritative source is named, and the date broke a tie only between equals.
- A person opened both original sources; Claude's summary was not the check.
- The battlecard shows one claim with a date, both claims with dates, or nothing, and the losing source stays in the ledger with the reason it lost.
- Every changeable claim has a recheck date, and pricing, legal and reputational claims have a named owner and a decision date.

## Where the Claude method runs out

The method above works with a spreadsheet ledger while someone can reopen every source on the recorded dates. A scheduled review can lag behind a pricing change. Until the next check, the ledger may disagree with the live page, and a seller might be the first person to flag the discrepancy. An ordinary, unscheduled chat does not monitor a page between requests. A paid Cowork scheduled task can run recurring research, but it still needs an accessible source or connector and a review step.

Steve's competitive-intelligence product flags battlecard items as out of date when new news or an uploaded resource may change them, and lets you review the update.

If a spreadsheet and a calendar reminder keep your recheck dates, stay with them. If missed rechecks have become a recurring seller complaint, review how [Steve's competitive-intelligence product](/products/competitive-intel) handles out-of-date battlecard items. Your team still decides what the card says about a competitor's price, and who signs off when the sources disagree.

## Frequently asked

### Does the newer source always win?

No. Recency decides only between sources of similar authority. A competitor's current pricing page outranks an older reseller listing, and a fresh forum comment does not outrank a dated official release note. Rank by authority first, then let the date break ties for prices and features.

### Does Claude detect contradictions in its research automatically?

Not as a documented feature. Claude's Research help pages, read September 16, 2026, describe multi-source research but do not promise automatic adjudication between two cited findings or automatic staleness detection. Ask for the comparison explicitly and open both sources yourself.

### What if the only source for a claim is old?

Treat the claim as unverified once its recheck date has passed, and reopen the source before repeating it. If the page has changed, the new page and the old row are a contradiction to resolve. If the page is gone, the claim stays unverified until you find a replacement.

### How does a seller see a disagreement on the battlecard?

As one line with both dates, for example the vendor price as of one date and a reseller figure from another month marked unconfirmed. If the sources are too weak to show, the card carries nothing for that claim, or says the claim is under review and not to be quoted.