How we make money

A comparison site that will not tell you how it is paid is asking you to trust it on the one point where its interests and yours can come apart. So: here is every way money can reach us, and the three things it can never buy.

Where we stand today

This site currently earns nothing. There are no paid links live on it, no advertising, no sponsorship and no paid placements, and no vendor pays us for coverage. We intend to add commission on some outbound “go to the tool” links, and when we do, the programmes will be listed here by name and every affected link will be labelled (paid link) on its face. This paragraph is generated from the same list that decides which links carry that label, so it cannot quietly fall out of date.

The three rules

Commission is paid after a reader has chosen, on a link whose destination was decided by a test result. That is the whole of the arrangement, and these three rules are what keep it harmless. They are not aspirations; each one is a property of how the site is built.

  • Commission never affects a score.

    A score is computed from the run receipts — the task, the answer, three judges' marks — by the same pipeline for every entrant. It is produced before any link to that product exists, and there is no field in the data a commercial relationship could be written into.

  • Commission never affects the order.

    Rankings sort on score. Where two things finish level, the tie-break is published and deterministic: fewest machine-checked rule breaches, then lowest measured cost per run. Both numbers come from the same receipts, so you can re-derive the order yourself.

  • Commission never affects who appears.

    Inclusion is earned by having a scored test run, not by having a programme we can join. Nothing is added because it pays and nothing is dropped because it does not.

Why this is a structure, not a promise

The scores exist before the links do. A verdict is produced by the testing pipeline from files you can open: the published task, the raw answer, three judges’ marks, a median, and a cap applied in code where an answer broke a checkable rule. Nothing in that chain has anywhere to put a commercial relationship, and the output is the same whether or not a programme exists. The protocol is published in full.

The tools we write about most cannot pay us at all. Six of the products our readers arrive for — ChatGPT, Claude, Gemini, Microsoft Copilot, NotebookLM and GPT Image — run no publisher commission programme of any kind. Neither do Midjourney or Suno. The pages that draw the most readers are, as a matter of arithmetic, the pages with no money behind them. That is an awkward commercial fact and a useful one for you: our attention follows what people ask about, because there is no alternative to follow.

We publish results that cost us money. A low score on a product we could earn from is published exactly as the protocol produced it. This is also why five programmes have been turned down outright — see below.

What a paid link looks like here

On the link itself

The words (paid link) immediately beside it. We use that phrasing because the US regulator endorses it by name and rejects the alternatives: consumers “might not understand that ‘affiliate link’ means that the person placing the link is getting paid”, and “commissionable link” is “probably not a clear disclosure”. The UK regulator reached the same conclusion about the word “affiliate” in a 2020 ruling.

At the top of the page

A plain block reading “Ad — paid links”, above the first commercial link, never in the footer and never behind a click. The ASA has ruled that a disclosure at the bottom of a page is unlikely to be sufficient, because a reader meets the link first. For the same reason it is never hedged into “may earn a commission” — the regulator names that construction as unacceptable.

In the markup

rel=“sponsored nofollow” on every paid link, which is what Google asks for on a link that exists for commercial reasons. A machine reading this site gets the same disclosure a person does.

What we refuse, and have refused

Before approaching any programme we read its actual agreement rather than its marketing page. Five were rejected outright because their terms restrict what a publisher may say — a non-disparagement clause, or a requirement that the vendor approve the material published alongside the link. We will not sign one, at any commission rate: a contract that makes an honest low score a breach would turn the only claim this site is built on into a lie. If a programme cannot live with us publishing the result our tests produce, it is not a revenue opportunity, it is a liability.

  • Payment for inclusion, a position, or a score — in any form, at any price.
  • “Featured”, “recommended” or “verified vendor” slots that a vendor can buy.
  • A bonus, tier or rate tied to where a product ranks.
  • A programme whose agreement restricts what we may publish about the product.
  • Vendor approval, preview or sign-off of anything before it goes live.
  • Removing or softening a verdict at a vendor’s request.
  • Sponsored content presented as a test, or a review written by anyone paid by the vendor.
  • Coupon, voucher, deal or cashback pages, and bidding on any brand term in paid search.

How to check any of this

Do not take the page above on trust. Four things on this site are checkable by a stranger in about five minutes:

  1. 1

    Open a score’s receipts. Every score on the site links the prompts, the raw answers and each judge’s verdict, with the date the run finished. If the answer does not support the mark, say so and we will publish the correction.

  2. 2

    Re-derive an order. Where two entrants tie on score, the published tie-break decides: fewest machine-checked rule breaches, then lowest measured cost. Both figures sit in the same receipts, so the order is reproducible rather than asserted.

  3. 3

    Check a price against its source. Every price carries the vendor page it came from and the date we read it. If the vendor has changed it since, the date tells you — and telling us gets it fixed.

  4. 4

    Compare this page against the site. The programme list above is generated from the same registry that labels the links. A commercial link on this site that is not labelled, or a labelled link to a programme not listed here, is a defect — report it and we will fix it and say so.

The standards we label against

In the UK, affiliate content is advertising and the CAP Code requires that marketing be obviously identifiable as such, before a reader engages with it. In the US, the FTC Endorsement Guides require a disclosure that is clear and conspicuous, as close to the recommendation as possible. Our labelling is built to both, and the wording above is taken from what each regulator has said is acceptable rather than from what other sites do.

One line from the FTC describes this site’s model exactly, and it is the test we hold ourselves to: if it is obvious what the ads are, and a commercial relationship does not affect what we review, what we say about it, or the placement and prominence of the review, then there is nothing to fix. The three rules at the top of this page are that sentence, enforced.

ASA guidance on affiliate marketing ↗ · FTC Endorsement Guides ↗

Questions people ask

Does anyone pay to appear on AI Intelligence?

No. No vendor pays for placement, position, ranking or inclusion, and none ever has. Products appear because they have a scored test run, and they rank on the score that run produced.

What does “(paid link)” mean on this site?

That we earn a commission if you buy through that link, at no extra cost to you. It is the wording the FTC endorses, in place of “affiliate link”, which its own research found readers do not understand as meaning the publisher is paid. Any page carrying one also carries an “Ad — paid links” notice above the first such link.

How do I know a commission did not change the ranking?

Open the receipts. Every score links the prompts, the raw answers and each judge's verdict, with dates. The score is computed from those files before any link exists, and the tie-break that decides equal scores is published and re-derivable from the same files.

Found something on this site that contradicts any of the above? That is a correction, and we publish those. Tell us — or read the corrections policy, the terms and what we collect.