You are picking between two messaging providers for a project that goes to a budget review on Friday. Your agent has already pulled a pricing record for each, and both look tidy: tiers, per-unit costs, a plan name for every price point. The comparison is nearly done. Before you paste the numbers into a slide, there is one question worth asking: who actually wrote those prices down?
That is the question a recent post on APIs.io puts front and center. The Plans collection lists 14,861 pricing documents, and the team can read 14,857 of them across 12,992 provider folders. Every one carries a method field, and the post argues that this field matters more than any price in the file.
Here is what it shows. 7,611 records were searched from the provider's own page. 4,725 of them, or 32%, were generated by the catalog itself. 7,727 are marked evidence_backed: true, and not one of the generated records is. The breakdown runs like this: searched 7,611, generated 4,725, estimated 1,808, a recorded absence 503, probed 121, no method field 58, and a small group of derived, unmarked or provider-published records that totals 31. Of those, only the searched and probed records are evidence backed, save for a single one in the last group. The other four links on the page are malformed, and the post says plainly that those are the catalog's own.
So if one of your two records is a generated plan shape, the numbers in it were never quoted by anyone. The post says 3,932 of the generated records follow the API Commons plan shape. They are useful as a map of what tiers exist, such as Webflow's 14 plans, but they are not a quote.
Now look at what a good record gives you. The Zoho CRM record transcribes what a program needs: api_credits_per_day and api_concurrency per tier. Free gets 5,000 credits a day and a concurrency of 5. Enterprise gets 50000 plus users times 1000 plus add-ons, capped at 5000000, and a concurrency of 20. The post calls that capacity planning, not marketing. Stripe's record is method searched, pricing_basis ad-valorem, with a date and a source URL, and its Standard plan reads 2.9% plus 30 cents per successful transaction for domestic cards.
Then there is Infobip, which the post praises for its honesty. Its note says the page uses interactive per-country calculators that show only zero-euro placeholders in the text, so the only explicit numbers captured are Viber minimum commitments of 175 and 115 euros. For you, that note is the difference between assuming the rest is free and knowing it was unreadable. The Datadog estimates file opens with the words ESTIMATES, not quotes, and says each estimate names the figure it came from and retires once the provider publishes that field.
Two more fields would change how you read your shortlist. The audience field tells you whether a record prices a business product or a developer API. Where it is set, 3,286 records price a business product and 1,860 price a developer API, and a subscription price is not the cost of an API call. And prices_published is set to true on 3,059 records and false on 1,273, while the other 10,525 have no value at all. The post notes that the false ones are a finding in their own right.
Here is the turn that makes the post worth trusting. The catalog does not hold itself up as the clean party. The work left, it says, is its own: the 4,725 generated records should be searched or probed against the provider's page, and prices_published should be filled in across the 10,525 records that lack it. Depth is also uneven. SignalWire has 64 plans in one record, while 4,626 records hold no plans at all, and 1,805 of those are estimate files that carry formulas in place of tiers.
Back to your Friday slide. Before you compare, filter on method and evidence_backed, and check audience so you are comparing like with like. Treat each pricing record as a claim with a source, and check the method before you trust the number.
Read the original: https://apis.io/2026/11/02/just-over-half-of-our-pricing-records-cite-evidence-and-32-percent-were-generated/