Skip to content
Measurement

The 48 percent ChatGPT lift for partner publishers counts pages that were already cited

Should PR chase publishers with OpenAI deals? What the 48 percent figure measures, where the lift stops, and how to pick targets from your own answers.

Obility Editorial · · 4 min read

The 48 percent figure averages over pages that were already cited

If you run PR and care about AI answers, you have probably been told to aim for publications that have a licensing deal with OpenAI. The number behind that advice comes from a study released in August 2026 by Press Ranger, a press release distribution company, and OtterlyAI, an AI search monitoring company. They matched 129.3 million citations collected in June 2026 across seven AI platforms against 91 known licensing agreements covering 314 publisher domains. On ChatGPT, pages from publishers with an OpenAI deal collected 10.2 citations each on average, against 6.9 for pages from publishers without one. That gap is the 48 percent.

Look at the denominator. The study’s own release describes the measure as citations per cited page, so a page enters the average only after some AI answer has cited it. The figure tells you that a partner page which does get cited tends to be cited more often. It does not tell you what share of partner articles are ever cited, and it says nothing about the article you might place next month. It also does not hold the type of page constant. By the release’s own count, best-of lists, buying guides and product reviews make up 46.9 percent of licensed publishers’ citations, and five groups (Future, Forbes, People Inc., Condé Nast and Hearst) take 69 percent of all citations to licensed publishers. A buying guide on a large review site would be reused often with or without a contract.

A deal makes a publisher eligible, and the counts show how little that guarantees

OpenAI’s announcements describe roughly the same arrangement each time. Its April 2024 post on the Financial Times partnership says ChatGPT users will see “select attributed summaries, quotes and rich links” to FT journalism in response to relevant queries. That is permission to display, not a promise of placement.

Ahrefs tested what that permission produces. It counted how many US ChatGPT answers in its Brand Radar index, 5.5 million answers in the November 2026 snapshot, cite each of 85 partner domains. Vogue appears in 42,108 and Car and Driver in 41,603. The Atlantic, which also has a deal, appears in 724. Investopedia, covered by the People Inc. deal, appears in 3,217, and Politico, covered by the Axel Springer deal, in 10. Ahrefs draws the same conclusion we do: a partnership makes a publisher eligible to appear and does not guarantee that it will.

Both datasets point the same way. The partner sites ChatGPT cites heavily publish reviews and buying advice, the kind of page that answers “which one should I pick”. A feature about your funding round in a prestigious partner title is a different kind of page, and nothing in either dataset suggests ChatGPT reaches for it when a buyer asks for a shortlist.

The lift sits on one engine

The Press Ranger and OtterlyAI study found the effect only on OpenAI’s own product. Publishers licensed by Google were cited in AI Overviews at a slightly lower rate than comparable unlicensed publishers, and Perplexity’s licensed publishers landed at parity on Perplexity. Publishers with OpenAI deals draw 57.9 percent of their AI citation volume from ChatGPT alone. If your buyers research in Perplexity, AI Mode or Copilot, a placement chosen for its ChatGPT effect is a narrow bet.

The same release contains the finding more useful to a B2B brand. News is 7.2 percent of all citations in the data, and across 16 US industries, niche and trade outlets, most of them unlicensed, carried the majority of news citations in 15. For a payroll tool or a security product, the publication most likely to sit under a buying question is a trade site no AI company has licensed.

A cited article is not a named brand

Every figure above counts how often a publisher’s URL appears among an answer’s sources. The question a brand cares about is different: does the answer name you. An article can be cited for one sentence about market size while the shortlist above it names four competitors. A placement helps only if the page says something the engine needs when it builds the list, such as what you do, who you suit and how you compare. A profile that never states your category in plain words gives the engine nothing to retrieve when a buyer asks a category question.

None of these studies followed a brand before and after a placement, either. The claim that a specific article moved a specific answer is untested in all of them.

Pick targets from your own answers, then check that the answer moves

Start from the questions your buyers ask, not from a partner list. Run them on each engine you care about and record every source each answer cites. Obility’s Answer Engine Insights keeps the answer, the brands it names and the sources it cites with each observation, so you can rank publications by how often they appear under your own buying questions, engine by engine. Which engines are covered is confirmed at the demo.

Then pitch the formats those cited pages already take. If the recurring source is a trade site’s annual roundup of tools in your category, a place in the next edition is worth more than a news mention in a bigger title. After the piece runs, keep asking the same questions and track two things separately: whether the new URL appears among the sources, and whether the answer names you. Give it weeks. Ahrefs found that only about a quarter of the pages ChatGPT cited for a prompt were cited again the next day, so a single check can show nothing or a false win.

Do not read a licensing deal as a lever you can pull through coverage. It is one fact about one engine. Let the sources in your own answers decide where the PR budget goes.

Share this article

Put the workflow into practice.

Bring your questions and content to an Obility demo.

Book a demo