The Prompt to Stop AI Making Things Up About Competitors

It is tempting to describe an invented competitor fact as the model being careless, or overconfident, or trained on bad data. Usually it is none of those. You asked a question, the material in front of it did not contain the answer, and every available response looked like a failure except producing one. So it produced one, in the same register as everything else it had said, because nothing distinguished the sentences it could stand up from the sentence it could not.

That framing matters because it points at the fix. You are not trying to make the model more truthful. You are trying to give it a second acceptable output for the case where the first one is unavailable, and then make that output specific enough to be useful.

The gap is invisible from your side

The reason this is worth a page of its own is that you cannot see which questions fell into a gap. The answer comes back complete. Nothing in it is hedged, because hedging would be a stylistic choice the model had no reason to make, and the invented sentence sits between two well-sourced ones inheriting their credibility. By the time it reaches a battlecard nobody remembers which claim came from where.

Six rules, and each one exists because of a specific way competitive analysis goes wrong. They are worth reading as a list of failures rather than as a list of instructions.

We picked a question deliberately: one where the source genuinely does not contain the answer, and where the near-miss is available. The vendor publishes a starting price for its top tier and does not say which billing period that figure assumes.

Notice what the first answer was not

It was not a hallucination in the way the word is normally used. There was no invented plan, no fictional number, nothing a search would contradict. The figure was correct and came from the page. What the model supplied was the sentence around it: a floor was reported as a price, and a billing basis was asserted where the page states none.

This is why advice to watch for hallucinations does not protect competitive work. Nobody is going to catch that by scanning for things that look made up, because nothing in it does. It is caught by having required, in advance, that any claim not on the page be labelled as not on the page.

The second answer is also worth reading for what it produced beyond the refusal. It quoted what the page does say, named the two specific things it does not, and ended with what would close the gap. A refusal in that shape is not a dead end, it is the next task.

The honest warning about this instruction is that people dislike its output at first. An ungrounded analysis reads as finished. It has an answer to everything, the paragraphs are even, and it can be pasted into a document without further thought. A grounded one arrives with visible holes, several claims marked as inferences, and a list at the end of things nobody could establish.

The holes were in the first version too. They were filled with material that looked exactly like the rest, which is the only difference between the two documents. Being able to see them is what lets you decide whether they matter for the decision in front of you, and usually most of them do not, which is a discovery you can only make once they are visible.

There is a second effect worth expecting: the gap list becomes your collection brief. After a few sessions the pattern of what the material never answers stops being random and starts naming the two or three sources you are not actually watching. The rest of the prompt library assumes this instruction is already in place, which is why almost none of the other prompts repeat it.

Grounding solves one failure completely and does nothing about the other. Once every claim is tied to a source you supplied, the output cannot contain a fact that was never true. It can still be full of facts that stopped being true, and the instruction has no way to tell, because a page you saved in March looks exactly like a page you saved this morning.

The two failures need opposite remedies, which is worth being explicit about. Invention is fixed by a rule in the conversation. Staleness is fixed by something outside the conversation noticing that a source changed, and no wording anywhere in the window can substitute for it. A model working faithfully from material captured six months ago will produce a beautifully sourced, entirely current-sounding, wrong analysis.

That is the failure a buyer catches rather than a colleague, because your side has no way to notice and theirs does. And it lands hardest on exactly the facts this instruction makes most quotable: the prices, the plan names and the capability claims, which are the ones a rep repeats word for word precisely because they were properly sourced.

Keeping the material current is monitoring rather than prompting. Flares watches competitor pricing, product and positioning pages continuously, so what goes into the window is this week’s and a grounded answer is current as well as sourced.

What it does not do is choose your sources for you. Deciding which pages, transcripts and exports belong in a competitive analysis is a judgement about what you are trying to decide, and it is the judgement this whole instruction exists to protect.