A method for auditing any shopping tool, including the one you are reading, and what we found auditing ours.
Every shopping tool has a business model, and the model determines the incentive. Affiliate commission on an outbound click rewards sending you somewhere. Commission on a completed purchase rewards sending you somewhere you actually buy. Placement fees reward showing a particular retailer regardless of either. Data licensing rewards observing what you do.
None of these are disqualifying, and a tool that earns nothing is usually a tool that will not exist in a year. Being unable to find out which model applies is the problem.
A tool that says it compares hundreds of retailers should be able to tell you which ones it saw for the search in front of you. A tool that promises the best available price should be able to show what it compared against. Where a claim cannot be inspected on a specific result, it is marketing rather than a finding.
The useful question is not whether a tool makes a large claim. It is whether the claim narrows when the evidence does. A tool that says the same confident thing about a search returning twenty retailers and a search returning one is not reporting; it is advertising.
Our browser extension used to describe itself as comparing more than fifty retailers, and its price badge said the figure included shipping. Both statements were untrue. The retailer count was aspirational rather than measured. The badge compared an item price scraped from the page you were on against an item price from elsewhere, with no shipping information on either side, and no way to obtain the shipping cost for the page you were currently viewing.
It also had no seller filter at all. On a Best Buy product page for a 299 dollar set of headphones, it once suggested a 159 dollar alternative from a liquidation seller as though the two were the same purchase. We measured that on 21 August 2026.
We rewrote the badge to state what it actually compares, which is item price against item price with delivery shown at checkout, and restricted it to new retail stock classified as such by the server. An offer the server has not classified is now treated as ineligible rather than assumed safe.
Publishing a fault you have already fixed is cheap. Publishing the mechanism that produced it is not, because it tells you what to look for in every other tool, including the ones we compete with.
The mechanism was this: a claim written during design, never re-checked against what the code actually did, and repeated in the marketing until it read as fact. That is not unusual. It is the default outcome unless something forces the claim and the behaviour back together. We now run an automated sweep over the whole codebase that fails the build when an unsupported claim appears in shipped copy.
Does it disclose how it earns, somewhere you can find without hunting? Does it distinguish a retailer from a marketplace seller? Does it tell you when its price was last checked? Does it ever say it does not know, or does every search produce a confident answer?
Does it change any link or referral credit on a page you visit, and does it say so? Does it work when you decline its permissions, or does it require access to every site you browse in order to show a coupon on one?
A tool that never expresses uncertainty is not more capable than one that does. It is less forthcoming.