Messaging Comparison Prompts for Competitor Claims

Most messaging comparisons are decided in the folder, before the prompt runs. What gets pasted is whatever was open. One vendor’s homepage. Another’s product page. A pricing page for the third. Your own positioning, taken from a deck rather than from anything a buyer can see.

The model compares all of it faithfully. That is the problem.

Each page is written for a different moment in somebody’s day. A homepage talks to a stranger who has not decided to care. A pricing page talks to someone who has. Compare across the two and you are comparing two audiences, then blaming the difference on the companies.

A headline and a feature bullet are not comparable

Put them side by side and the headline looks vague while the bullet looks narrow. Neither of those is a fact about the vendor. It is a fact about the job each sentence was hired to do. Ask a model to compare them and it will describe the gap in the language of positioning, because that is the language your prompt handed it.

The same URL is not the same page everywhere

On 21 September 2026 we opened the English URL of one of the two vendors in the run below. It came back in French, headline included. Nothing in the pasted text says so. Anyone building a comparison from two tabs that morning was comparing an English page with a French one, and any difference in emphasis was partly a translation.

Two inputs, and the first is a discipline rather than a document. Pick the page type first. Then go and collect that one page type from everybody, including yourself.

If a vendor genuinely has no version of that page, record the absence. Do not fill it with the nearest thing. A competitor with no pricing page has told you something. A competitor whose pricing page you swapped for a product page has told you nothing, and will now appear to disagree with everyone.

Four prompts. The first produces the three lists everybody wants. The second is the version that survives a check, because it works one page type at a time and refuses to substitute. The third is a tester rather than a generator: point it at a comparison you already have and most of the contested list will not come back. The fourth reads the same page across dates instead of across vendors.

Step one of the third prompt is the whole instrument. Write, in one noun phrase, the single thing both claims are about. Setup time. Reporting depth. Who the product is for. If you cannot write that phrase, the two quotes are not talking to each other, and the pair should be deleted rather than argued over in a meeting six weeks later.

A contested entry is not a quote. It is an assertion that two quotes are in conflict, and nobody sourced that assertion. Both halves can be word-for-word accurate. Both can be correctly attributed and correctly dated. The pair can still be wrong, because the two sentences were never about the same thing.

That is what makes this failure stick. An invented quote is caught by searching the page for it. A manufactured contradiction passes every check anyone thinks to run, because each half passes on its own. It then travels into a differentiation matrix and onto a slide, where somebody states that two companies disagree about something they have never both discussed.

A related habit sits one level up, and it belongs to a competitor SWOT, where a named box gets filled whether or not the material supports it. The difference is what gets manufactured. A half-empty quadrant produces a weak claim about one company. A manufactured pair produces a confident claim about the relationship between two, and the relationship is the thing you ran the comparison to find.

We asked for the contested claims between two CRM vendors with nothing in the window. Then we pasted both live pages in and asked again.

Read the three pairs as pairs, not as statements. Two vendors both claiming to be easy are agreeing. Two vendors naming different buyers are describing different books of business. Only the first pair is even shaped like a contradiction, and that is the one the page contradicts.

When two competitors both claim focus

The breadth claim belongs to the platform, not to the product page in the comparison. On the day we read it, that page led with a sentence about doing less. Two things got merged: a company’s best-known position, which lives in years of writing about it, and the copy currently on one product page. The first is easy to recall. The second is what the buyer is reading.

The grounded run came back with an empty contested list and a shared list holding the claim each vendor treats as its own. That is an uncomfortable result and a useful one. On the page where both companies spend the most care, a buyer comparing them in one sitting hears the same promise twice.

The list everyone reads first is the one a model is least able to produce. Being unique is a claim about a whole market. What the model can see is your folder, on one page type, on one day.

So read that list literally. It means: no other vendor put this claim on this page type this week. That is a much smaller statement than “only we do this”.

Two things follow, and both are practical. A claim that lives in a competitor’s documentation but not on their homepage will land in your unique column, which is why the disciplined prompt asks what page each block came from. And a claim that is unique because nobody else bothered to write it down is not a differentiator at all. It is category wallpaper with a two-week head start.

An absence in a competitor’s own documentation is a separate question with its own answer, and a feature comparison prompt is where the four meanings of an empty cell get separated. The test here is cheaper. For each item in your unique column, name the page a competitor would have put it on. Then go and look at that page before the claim reaches a slide.

A one-day comparison tells you where two vendors stand. It does not tell you where either of them is heading, and that is usually the more useful half.

The fix is to compare the same page across dates instead of across vendors. Web archives hold dated snapshots of most public pages, so you can read one vendor’s homepage at four points over two years and watch the sentences change. Do that for each vendor, then lay the timelines side by side.

Three things show up that a snapshot cannot give you.

Two cautions, and neither is fatal. Archive coverage is uneven, so a missing snapshot is not a missing change: date your findings to the snapshots you actually have. And an archive records what a crawler saw, which may be a different country’s version of the page from the one you get today.

The direction of travel is the output worth keeping. A position is arguable. A vendor who has added the same kind of language in three consecutive rewrites is not arguable, and you found it before anybody had to lose a deal to notice.

Marketing copy is the fastest-moving public thing a company owns, and the only one with no changelog. A pricing page moves when packaging moves. A homepage moves on a campaign cycle. Neither announces it. So a comparison is true of a date rather than of a pair of companies, which is why the capture date belongs beside every quote rather than at the top of the document.

The moment worth catching is not a rewrite. It is a competitor adopting a sentence you were using. Your differentiator becomes a shared claim without anything on your side changing, and nobody notices from the inside, because your own copy still reads exactly as it did.

One limit does not move, and it is worth being plain about. What a competitor claims is public. What they can prove is not. A comparison built from live pages maps the claims and stops there. Getting past that needs what they pay to promote and what their customers say, which are separate jobs with separate evidence.

Flares watches the pages a comparison is built from and stamps a date on every sentence that moves. The next run starts from this week’s wording instead of whatever somebody collected last time. Deciding what a new sentence means for yours is still a judgement call.

Which is the argument for keeping the page label attached to every claim you carry forward. A quote that still names the page it sat on is one search away from being confirmed. A quote stored on its own turns into something a company is said to claim, and nobody can find where it was ever said.