Perplexity Comet Plus Publishers: What Brands Should Pitch

Perplexity's Comet Plus pays participating publishers from subscription revenue when Comet users visit their articles, when Perplexity cites them and when Comet's agent uses them. It pays no brand for being mentioned in those articles, and the other engines' publisher programs work the same way. What the deals do tell you is which outlets the engines pay to read, and that belongs on your pitch list.

What Comet Plus pays publishers for

Comet Plus is a Perplexity subscription tier, and a share of Perplexity's subscription revenue, including Comet Plus, goes to participating publishers. It is weighted by use: people reading in the Comet browser, Perplexity citing the content in answers, and Comet's assistant using it to finish tasks. Perplexity keeps a share for compute. Brands are not part of it.

Perplexity announced it on 25 August 2025 as a $5 standalone subscription, included in its Pro and Max plans, and describes the payout as allocated by "three types of internet traffic: human visits, search citations, and agent actions." Its own post says it distributes the revenue "minus a small portion" for compute. It does not print a number.

The number comes from interviews. Digiday reported that revenue from Perplexity's paid subscriptions, Comet Plus among them, is pooled, and that Perplexity keeps 20% and 80% goes to participating publishers. Asked whether the three kinds of use are weighted differently, Perplexity's head of publisher partnerships said "The details will be very specific to our partners". So nobody outside those contracts knows what a citation is worth.

Separately, Digiday reported an initial publisher pool of $42.5 million, and Engadget, citing The Wall Street Journal, said the $42.5 million revenue pool will expand over time.

Perplexity's 2024 program was different, because it was funded by ads a brand could buy. The Publishers' Program shared that ad revenue with outlets referenced in an answer carrying an ad. Those ads are gone: The Keyword, reporting the Financial Times in February 2026, said ads had been halted with "no plans to pursue them further."

CNN was on Perplexity's list, then off it

Perplexity named its launch partners on 1 October 2025, and an archived copy of that page from 4 October 2025 lists CNN first, as did PPC Land's report of 3 October 2025. When we read the live page on 16 October 2026, it named The Washington Post, the Los Angeles Times, Fortune, Le Monde, Le Figaro and Condé Nast, whose titles include Allure and Vogue. CNN was gone.

Perplexity has not said why. What is on record: on 28 May 2026 CNN sued Perplexity, and CNN's own report on the suit says the network sought a content deal "but did not agree on terms." Several pages that rank for this topic still list CNN as a partner, so check each name on your list against the program's current page.

How the other engines pay publishers, compared

Four other companies pay publishers for AI use, each a different way: Google per contribution in a pilot, Microsoft through a marketplace pilot, Amazon and OpenAI through licensing deals. In all five programs the money goes to the outlet that owns the content, and none offers a brand a way in.

On 16 October 2026 we read each company's own announcement where one exists (Perplexity, Microsoft) and the reporting on the rest (Digiday, Search Engine Journal, TNW, LLM Pulse). Amazon's and OpenAI's deal terms are not published.

Google pays for influence on the answer. Search Engine Journal quoted the pilot's help text: content that only confirms facts or gets linked after the answer is written is excluded, because those instances "don’t qualify for AI contribution." Google has not named the participants. TNW, citing The Information, reported about 100 sites, with several small ones paid less than $1,000 over several months, one early participant earning more than $1m a year, and niche sites such as anime and gaming earning more. TNW says it has not verified the figures.

Microsoft pays per use through a marketplace it still calls a pilot. Its announcement of 3 February 2026 says publishers "will be paid on delivered value" and names the publishers that helped design it, without saying which supply content. One line in Microsoft's Publisher Content Marketplace post speaks to stores directly: for "a considered purchase", trusted sources help the AI give guidance "shaping the ultimate decision."

Amazon licenses lifestyle and shopping content. Digiday reported on 10 July 2025 that Condé Nast and Hearst signed multi-year deals for Rufus, Amazon's shopping assistant, with Hearst naming Good Housekeeping, Cosmopolitan and Harper's Bazaar. As of that report, Amazon had Alexa+ deals with more than 200 publishers.

OpenAI licenses by contract. Al Jazeera reported the Condé Nast deal in August 2024, giving OpenAI "permission to display content" in ChatGPT. LLM Pulse's tracker, updated 12 October 2026, still lists Condé Nast's 2024 deal (training and display) and Hearst's (display with attribution) among more than 20 news publishers. OpenAI began testing ads for free-tier ChatGPT users (reported February 2026), through a separate ads product that our ChatGPT product feed guide covers.

What money buys a brand, and what it does not

A brand's money buys an ad, a labeled sponsored post or a commission on affiliate sales. It cannot buy a place in any of these publisher programs, or an editor's verdict. Engines pay publishers because they treat editorial coverage as independent evidence. Once a brand pays for that coverage, it is an ad, and an ad is not independent.

Affiliate roundups sit between paid and earned. Before you pitch a shopping editor, ask which affiliate network the outlet's commerce team works with, and join it if you are not on it. Google's spam policies treat affiliate content as acceptable when "affiliate links are clearly marked", so expect the disclosure.

Keep review samples clean. Google's spam policies list "Sending someone a product in exchange for them writing about it and including a link" as link buying, so never make a link the condition of a sample. If you buy a sponsored post, write the label into the contract: an unlabeled paid link breaks those policies and, if it is spotted, discredits the outlet's other coverage of you.

As I wrote about seeding community threads: "you can't control it and you absolutely should not fake it. Astroturfing gets detected, gets you banned, and poisons the exact signal you were trying to build."

The honest frame is the one on our homepage: "Nobody can edit an AI model's output. What you can change is the evidence it draws on: the accuracy of your entity data, the structure of your content, and what independent third parties publish about you." Earned coverage is that third input. Your own entity data is the first.

Which outlets are worth pitching

Pitch the outlets that already cover your category and already appear when buyers ask an AI engine about it. Program membership comes third, as a tiebreaker: an outlet named in several engines' programs may be read by more of them, though no engine has said a deal raises citations. Ask these five questions in order.

1. Does it publish roundups, reviews or comparisons in your category? A news site that never covers moisturizers is a poor pitch, whatever it is licensed for.
2. Does it show up in your buyer prompts? Run your category questions in each engine you care about and log the cited domains. Our GEO Measurement Protocol sets out how to run that without fooling yourself.
3. How many programs name its publisher, and how? Check the table above, then the program's own page. Count a confirmed deal higher than a co-design credit or a vendor tracker's entry, and write down the date you checked.
4. Is the coverage editorial? A labeled sponsored slot is a media buy, judged on its readers.
5. Can you give the editor checkable facts? Send a one-page fact sheet with price, sizes, key ingredients, shipping regions and a test result, worded exactly as on your product page, so whatever the outlet prints matches your entity data.

Ask question 2 before question 3. Question 2 is evidence from the engines themselves, while question 3 is an inference: a license gives an engine access to an outlet's content, and none of these companies has published that a deal changes how often a partner is cited. Different engines also lean on different sources, as our comparison of how each AI search engine finds stores shows, so log each one separately:

A worked example: a skincare brand with three targets

Take a skincare brand weighing three pitches for a new moisturizer: Allure, Good Housekeeping and an independent skincare blog that ranks well for ingredient questions. The program evidence comes from the comparison table. The brand is hypothetical, and the example includes no citation data: that is the log above, which you fill in for your own category.

• Allure belongs to Condé Nast: three confirmed deals (Allure is listed on Perplexity's Comet Plus page, plus OpenAI and Amazon's Rufus), and a co-design credit on Microsoft's marketplace.
• Good Housekeeping belongs to Hearst: one confirmed deal (Hearst named it in the Rufus deal), one recorded by a vendor tracker (OpenAI), and the same Microsoft co-design credit.
• The independent blog appears in none of the published lists. Google does not name its pilot sites, so it could be in that one without anyone outside knowing.

No outlet can score in all five, because Google does not publish its pilot list. If all three pass questions 1 and 2, Allure goes first and Good Housekeeping second. If the blog shows up in your Perplexity and ChatGPT answers and Allure does not, the blog goes first. What the engines already cite outranks what they have licensed.

Earn the mention, then measure it

Editorial coverage takes months to win and cannot be bought outright. Engines treat it as independent, which is what makes it worth more than an ad. Give editors two things you control, a product to test and a fact sheet they can trust, then check prompt by prompt whether the article starts showing up in answers.

What goes in the pitch

Pick a roundup the outlet already runs that your log shows engines citing, and name its URL in the first line. Keep the pitch to four lines: which roundup, what your product does differently for that roundup's reader, the one fact that proves it (a test result or an ingredient concentration), and the offer of a sample. Attach the fact sheet from question 5 so the editor can copy your details exactly. Send samples only on request, with no strings attached. Follow up once, after seven days, then leave it until the next update of that roundup.

As I argued in GEO vs SEO: "The work that moves AI citations doesn't happen on your website." It happens in outlets like these.

Start with your buyer prompts. Fill in the log for each engine, run the five questions on the outlets it surfaces, and send three pitches. If you want help, our GEO service starts with an AI visibility audit that baselines where you appear and where competitors appear instead, and its citation-building work goes after the roundups and comparison pages that feed your category.