How free credits are checked and sourced

This page sets out how buildaicurrent gathers, checks, and orders the figures on the site. The goal is that every number can be traced back to an official page, read on a stated date, with nothing invented and nothing borrowed from secondhand sources.

Core rules

Official pages only. Every figure comes from a first-party page: the provider's own pricing, docs, or legal terms. Blog posts, roundups, and aggregator sites are not sources.
Every number has a source and a date. Each content page lists the official pages it relies on and carries a "last checked" date, which is the day those pages were actually read.
"Not stated officially" is fail-closed. If an official page does not state an amount, expiry, or payment rule, the site says "Not stated officially" rather than filling the gap with a guess or a figure copied from elsewhere.
"Not tested yet" is honest. Anything the writer has not personally run is marked "Not tested yet". The site would rather publish a known gap than a fabricated experience.
Correcting false numbers is core work. When a figure circulates widely but is absent from the official pages, the site says so explicitly and dates the check. Pointing out that gap is part of the job, not an aside; thecorrections index collects every such claim in one place.

Grant types

Each credit entry is classified by one grant type. The definitions used on the site are:

Signup credit: One-time signup credit
Monthly free tier: Monthly free tier
Top-up bonus: Top-up bonus
Always-free tier: Always-free tier, no expiring credit
Free trial: Time-limited free trial
No free credit: No free credit offered

Recheck policy

Checking is done by one writer. The "last checked" dates are intentionally staggered across pages rather than refreshed all at once, so the dataset ages evenly and no single batch hides a stale page. The date on this methodology page is 2026-09-18. If a page here disagrees with the official page it links to, trust the official page and treat the site figure as stale; the "last checked" date shows how stale.

The daily source check

A second, narrower check runs every morning, and it is not the same thing as the recheck above. It takes the figure each entry publishes and looks for that figure on the sources that entry already cites. If the number is gone from every one of them, the check fails and the entry gets read by a person. It caught Beam withdrawing its $30 per month allowance on 2026-09-09, the day after the previous run had found it still in place.

What it cannot do is more important than what it can. It matches a string, so it only covers entries whose amount is specific enough to search for, currently 29 of 55; entries that state no credit, or state one in words, are invisible to it. It cannot notice a changed condition, a new card requirement, or an offer that still shows the same number under different terms. Only a person reading the page catches those, which is what the monthly recheck is for.

The date in the masthead is this daily check, not the monthly read, which is why it moves most mornings. It means "the published figures were matched against their sources on this date", never "every entry was reread by a person on this date"; the per-entry "last checked" date is the one that means the second thing, and it is the authoritative one if the two ever disagree. Every run, including the days it found nothing and any day it did not run at all, is listed on the changelog. If the job ever stops, the masthead stops claiming a daily cadence rather than freezing on its last good date.

Quoted wording

Quotation marks on this site mean the words inside them are the provider's, verbatim. Not a tidied version, not a shortened one. The distinction matters because a paraphrase that keeps the meaning can still lose the condition: a sentence that reads "no free trial" in a quote may in fact say no for two products where it used to say no for one.

A script checks this the way the daily job checks figures: it takes every quoted phrase in an entry and looks for it on the pages that entry cites. It is not a gate and does not run on a schedule, because most of what it flags is innocent, a page that renders its text with JavaScript, or a claim the entry quotes precisely because no official page makes it. It runs before a recheck round, and a person reads the list. On its first full pass over the entries no daily check can watch, it flagged 22 phrases, of which two were real: Salad had rewritten its no-free-trial answer to cover a second product and dropped a transcription allowance, and Beam had replaced its terms of service, removing the clause that prohibited benchmarking. Four more were this site's own paraphrases sitting inside quotation marks, which have been corrected.

Ranking and money

There is no unpublished composite score behind the order of entries. Tables and indexes are sorted by fields that are shown openly, such as the last checked date or the product name. No provider has paid for placement, and product links go to official pages. If affiliate relationships are ever added, they will never change the order of entries, and they will be disclosed on the disclosure page and inside the affected article. Today there are none.

Open data

The full credits dataset is published as machine-readable files so anyone can reuse or audit it: JSON, CSV. Both are generated at build time from the same entries shown on the credits index, with drafts excluded. The dataset is licensed under the CC BY 4.0 license; the JSON file states this in its license field.

Schema version and field stability

The JSON payload carries a schemaVersion field, currently 1, so a consumer can pin the shape it was written against. Within a major schema version, fields are never removed and never repurposed. New fields may be added, so a reader that ignores unknown fields keeps working, and new CSV columns are appended at the end for the same reason. Removing or repurposing a field means a new major version, and the version number changes with it.

The machine-readable endpoints are the data at /credits.json and /credits.csv, plus a feed of dataset changes at /changelog/changes.xml. That feed repeats the schema version and the policy above in its channel metadata, and gives every change a machine-readable kind, entry id, official source URL, and check date. It covers the same rechecks recorded on the changelog, including the ones that found no change.