The Best Competitor Pricing Comparison Prompts

Most pricing comparisons go wrong before any arithmetic happens. Two vendors publish numbers, somebody puts them side by side, and the table implies the two figures describe the same purchase. They rarely do. One assumes annual billing and the other monthly, one counts a platform fee the other folds into the seat price, and one of them charges for onboarding in a line the table never reached.

So the first decision is not which prompt to use. It is what configuration you are comparing at: how many seats, over what term, billed how, in which currency, with which capabilities required. Write that line down before you run anything. Every prompt below produces a different answer for a different configuration, and a comparison that does not state its own is not reproducible by the colleague who questions it.

The word price hides two numbers

Almost every software pricing page carries a billing toggle. The two figures behind it are not close to each other, and the gap is widest exactly where buyers start. HubSpot’s Sales Hub page in September 2026 puts Starter at $7 per seat per month billed annually and $20 billed monthly, which is the same plan described by two numbers nearly three times apart.

A table with one price column has therefore already made a decision and not recorded it. The prompts here refuse to let the model make that decision, which is why the first one spends a rule on it rather than on tone or persona. Price intelligence as a discipline is largely the practice of keeping these distinctions straight at scale.

Run them in order. Each one consumes the output of the one before it, so the table built first becomes the input to the cost model and the tier analysis rather than being rebuilt three times with small differences.

The fourth is the one people skip and the one that most often changes a decision. A buyer comparing two vendors is rarely stopped by a price; they are stopped by something neither page told them, and a list of those gaps is a better brief for a sales conversation than the comparison itself.

Pricing is the single worst subject to ask a model about from memory, because the answer is numeric, confident and checkable, and nothing about it signals its age. Three things belong in the context window first.

We ran the same question twice against one vendor whose prices are published openly, once with nothing supplied and once with the page saved that morning. The point of printing both is that the ungrounded answer is not obviously bad. It is the kind of answer somebody checks, nods at, and puts in a deck.

The failure was not invention

Every tier name came back correct. The onboarding charges came back correct. What went wrong was subtler: each per-seat figure quoted was the monthly-billing price, presented without saying so. Against an annual comparison that overstates the entry tier by roughly 186 per cent, and it does it consistently in the same direction, so nothing inside the answer looks inconsistent enough to query.

This is why “check the AI for hallucinations” is not a workable instruction for pricing work. Nobody is scanning for a fabricated plan name. The thing to check is whether each figure knows which billing period it belongs to, and the only way to check that is to have the page open.

What a grounded answer still cannot fix

The second run is accurate and its accuracy has a shelf life measured from the moment the page was captured. That is a genuine improvement, because a dated claim can be re-checked and an undated one cannot, but it is worth being precise about what changed: the risk moved from being wrong to being old. Those need different remedies, and only one of them is solved by a better prompt.

Three checks, in this order, because they are ordered by what costs you the room.

Then give the output somewhere to live. A comparison that stays in a chat window is re-derived by the next person who needs it, slightly differently. The competitive pricing matrix takes the table with its columns intact and exports it, which matters mainly because it gives every figure a cell to sit in beside the date it was captured.

The comparison you just built is correct and it has already started ageing. What makes pricing different from most competitive work is the shape of the decay: it is not gradual, it is a step, and it happens on a day nobody announces.

A price rise is the easy case, because it is visible the next time anybody looks. The expensive case is a repackaging. A vendor moves one capability from the middle tier to the top one and every published figure on their page stays exactly the same, so a comparison built on those figures remains arithmetically perfect while describing a product that no longer exists at that price. Nothing in the output looks stale, which is precisely the problem.

Being late on that has a specific cost, and it is not embarrassment. It is a rep quoting a tier that was reorganised last quarter, in front of a buyer who has the competitor’s page open on a second screen, and losing the argument about everything else in the comparison at the same time.

Watching for it is ordinary competitive monitoring rather than anything a prompt can do. Flares tracks competitor pricing pages and what each tier contains, so a repackaging arrives as a notification rather than as a surprise in a deal, and the pages you paste into this prompt are the current ones.

What monitoring does not do is decide what the change means for your price. Whether a competitor moving a capability upmarket is an opening or a threat depends on your own packaging and your own buyers, and that is the judgement this prompt was always there to support rather than to make.