Competitor Positioning Prompts That Quote the Page

Ask for a summary of how a competitor positions itself and you get a fair description of what the company is about. It will usually be accurate. It will also be unusable, because positioning is not a substance that can be described; it is a small number of specific sentences that somebody argued over, and those sentences are the thing a buyer repeats back to your rep.

The difference shows up the moment the output leaves the document. A rep who has been told a competitor is “an easy-to-use all-in-one platform for growing teams” has been told nothing they can use in a room. A rep who knows the competitor’s page says six words, and knows which six, can hear those words coming and answer them.

Near-quotes are worse than no quotes

There is a failure specific to this job and it is quiet. A model asked to quote will sometimes produce a sentence that is almost the page: the same idea, the same shape, two words different. It passes every internal review because everyone reads it for meaning. It fails in front of a buyer, who has read the actual page, and a competitor who can point out that they never said that gets to dismiss everything else you claimed in the same breath. Checking quotes by searching the saved text takes a minute and is the only defence.

Three sets of pages, and the second is the one that gets skipped.

Save text rather than take screenshots. A screenshot captures one viewport, crops the footnotes and cannot be searched, which defeats the one verification step this job actually has.

The first extracts. The second grades what it extracted. The third reads what is not there. The fourth is the only one that needs your own pages, and it is the one that turns a teardown into a decision.

Every one of them ends with the same escape hatch: write “not stated” rather than filling the gap. On this job that instruction does more work than anywhere else in the library, because marketing pages are written to imply things they do not say, and a model reading them helpfully will complete the implication on the author’s behalf.

We asked about a real product, then opened its live page. The answer invented nothing, and it still produced two things you would not want to act on.

The third row is the interesting one. Given nothing, the model produced a comparative framing that sounded right and that the page does not make. Given the page and told to work from it, the same model recorded the absence instead. Nothing about the prompt quality changed between those two answers. What changed was whether there was a document in front of it.

Once the wording is out, the next question is which of it carries weight. The test is operational rather than philosophical: what could a sceptical buyer actually check?

Sorting a page this way changes what you do next. Assertions are answerable directly, usually by asking the buyer what they would want to see. Proof has to be matched with proof of your own, and a competitor with eight named customers in your segment is making an argument that no amount of positioning language answers.

The absences are evidence too, and they are not conclusions

The third prompt above reads the navigation for what it never mentions: roles with no page, industries never named, objections the copy never approaches. This is genuinely informative and it is the easiest place on this page to overreach. A missing use case may be a retreat, an unwritten page, or content sitting somewhere you did not save. Record the absence, write down what would distinguish the explanations, and go and find out. An absence presented as a strategic choice is a story, and stories about competitors have a way of surviving longer than the evidence for them.

A teardown is a photograph. It is accurate about a set of pages on the date at the top of your capture, and from that moment it begins to describe a website that no longer exists in quite that form.

Positioning changes are the worst case for this, because they are both consequential and silent. Nobody announces a new homepage. A promise gets rewritten, a buyer segment disappears from the navigation, a proof point is added or quietly dropped, and none of it produces a notification. The first sign is usually a rep reporting that a buyer said something the battlecard does not cover.

Flares watches competitor homepages, product pages and pricing pages and dates every change, so a rewritten promise reaches you when it is rewritten rather than the next time somebody decides to look. The reading of what the change means stays yours, and it is the part worth your time.

It is also worth saying what monitoring does not settle. Knowing that a competitor changed their headline does not tell you whether to change yours, and copying a repositioning is how two companies end up making the same promise to the same buyer with nothing to choose between them.