Unsourced
Analysis

AI engines are starting to pay the publishers they cite. Can you prove you are one?

For two years the story of AI search has been extraction: assistants read your work, answer the question, and keep the click. That story is starting to change, and the change has a price tag attached.

Published 13 July 2026 · Unsourced Research · All research

The last two years: traffic left, and it did not come back

The numbers on zero-click search are no longer subtle. Google's zero-click rate climbed sharply in the year after AI Overviews launched, and organic traffic to news publishers fell by hundreds of millions of monthly visits over the same window.

If that were the whole story, it would be a slow bleed. But something moved the other way too: ChatGPT referrals to publishers grew many times over year on year. The front door is being rebuilt. The traffic that survives increasingly arrives through an AI answer, and it tends to convert better than cold search because those visitors are further down the funnel. Fewer visitors, warmer ones.

The new wrinkle: engines paying for what they cite

Here is the development worth watching. Perplexity's Comet Plus programme reportedly sets aside a $42.5M pool for publishers, split 80/20 in the publishers' favour, allocated according to whose content gets cited in AI answers (Digiday has the mechanics). Set the exact figures aside. It is the first serious attempt to align an engine's economics with the sources it leans on.

For a decade, being "the source" was a cost you absorbed so you could win the click. In this model, being the cited source is the revenue event itself. That reframes a loose question, "are we visible in AI?", into a sharp one: can you prove, per engine, that your site was the cited source, with evidence someone else would accept?

Why a visibility score will not survive contact with an invoice

Most AI-visibility tools hand you a single number and a trend line. That was tolerable while the stakes were reputational. It stops being tolerable the moment citations map to payouts or a rights claim, because a score is an estimate you cannot audit, and the divergence between engines is enormous.

We know how large, because we measured it: ask seven leading assistants the same question and they agree on a source only about 16% of the time. There is no single "AI" to be visible in. Being cited by ChatGPT tells you almost nothing about whether Perplexity or Gemini cited you, and if Perplexity is the one writing cheques, Perplexity's citation record is the one that matters, with dates and the exact answer attached. We wrote up why a score is a guess and a receipt is not.

What proof actually looks like

This is the gap Unsourced was built for. Instead of scoring you, it captures the evidence:

None of that is a number you take on faith. It is a record you can open, hand to a partner, or attach to a claim.

What to do this week

You do not need to wait for a cheque to act:

The extraction era rewarded whoever ranked. The era starting now rewards whoever can prove they were the source, cleanly, per engine, with the receipt in hand.

Common questions

Are AI engines really paying publishers for citations?

It is early, but yes, the model is appearing. Perplexity's Comet Plus programme reportedly sets aside a $42.5M pool for publishers, split 80/20 in their favour, allocated in part according to whose content gets cited in AI answers. Set the exact figures aside; the direction is what matters. Being the cited source is starting to be the revenue event rather than a cost you absorb.

Why will an AI visibility score not be enough?

A score is a single blended estimate you cannot audit. The moment citations map to payouts, licensing or a rights claim, you need to prove, per engine, that your specific page was the cited source, with the date and the exact answer attached. A number that went up or down does not do that.

The engines disagree — does being cited by one tell me about the others?

Almost nothing. We asked seven leading assistants the same question and they agreed on a source only about 16% of the time. There is no single "AI" to be visible in. If Perplexity is the one writing cheques, Perplexity’s citation record is the one that matters, specifically.

What should I do this week?

Run your domain through the free AI Visibility Check to see where you actually stand across five models. Watch the divergence between engines rather than any single score. And start keeping dated, per-engine evidence now, so you are ready when "cited = paid" becomes normal rather than reconstructing it later.

Related

Free AI Visibility Check

Ask five AI models your real questions and see who they name.

Read it →

Citation receipts vs an inferred score

Why a blended score cannot back a licensing or rights claim.

Read it →

A score is a guess. Ask for the receipt.

What auditable citation evidence actually looks like.

Read it →

See the evidence for your own site

Unsourced captures which AI assistants cite you, proves which crawlers really fetched your pages, and re-checks after you act — evidence, not a score.

© Unsourced — the evidence layer for AI search.