Perplexity’s Ad-Free Bet vs OpenAI’s Ad Model: A Publisher’s Response

TL;DR

•  The usual comparison — Perplexity’s trust-first ad-free bet versus OpenAI’s ad-funded scale bet — treats “ads: yes or no” as the decision. For a publisher, that is the wrong variable.

•  A platform’s stance toward your content is downstream of who pays the platform, not of what it says in its principles. Read the revenue flow and you can predict the behaviour.

•  Follow the money and the intuition inverts: the ad-free platform (Perplexity) shares 80% of a subscription pool with publishers; the ad platform (OpenAI) shares nothing per citation and monetises the answer itself.

•  The Revenue-Flow Ledger traces who pays → what is sold → where your content sits → what the platform is therefore incentivised to do to your citation. The Publisher’s Position Test turns that reading into one of four postures per platform.

•  But alignment is directional, not generous: $42.5M across 2,400 publishers is rounding-error money, and Perplexity is being sued by the same outlets it pays. Neither model is cleanly good for publishers.

•  The response is not to pick the winning business model. It is to own the one asset both bets leave un-buyable — the earned citation — and position to capture its value under whichever model wins in your market.

The comparison everyone is running — and why it answers the wrong question

Open any AI-search trade post from mid-2026 and you will find the same head-to-head. On one side, OpenAI, which started testing ads in ChatGPT in February 2026 and reached roughly $100M in annualised revenue within weeks — the fastest ramp for a new ad surface since TikTok, per DigitalApplied’s tracking. On the other, Perplexity, which in the very same month did the opposite: it discontinued advertising entirely and moved to a subscription-only model, and reportedly saw revenue jump about 50% the following month, per Stan Ventures’ July 2026 state-of-Perplexity report. Two companies, one month, opposite bets. The trade posts frame the takeaway as a media-buying question: which ad platform deserves your budget?

That is a real question for an advertiser. It is the wrong question for a publisher, and it is the wrong question for anyone whose job is to earn citations rather than buy placements. The ad decision — run ads, don’t run ads — is a symptom. It sits downstream of something that actually determines how each platform will treat the content you produce over the next three years: who the platform’s paying customer is.

Here is the counterparty claim this article dismantles, stated as fairly as its holders would put it: “Perplexity removed itself from the ad buy, so the strategic choice is settled — advertise on ChatGPT and Google, and treat Perplexity’s ad-free stance as a nice trust story with no bearing on where my money or effort goes.” That is coherent for a media buyer with a next-quarter plan. It falls apart the moment the reader is a publisher deciding how to invest in being cited, because for a publisher the ad-free-versus-ads split is not about ad inventory at all. It is a tell about the platform’s incentives toward the people who supply its answers.

This piece builds the tool for reading that tell. It is a fair comparison, not a cheerleading exercise: both models have a genuine case, and — as we will see — both leave publishers worse off than the old web in ways their marketing obscures. If you want the ground-level mechanics of how an earned mention beats a bought one inside an answer, that argument lives in our earlier work on what backlinks and citations actually do. Here the altitude is higher: the business model itself, and what it predicts.

Follow the money, not the mission statement

Start with a principle that will do most of the work in this article: a platform’s public stance toward publishers is a downstream expression of who pays it. The values track the customer. When a company’s revenue comes from a group, its product drifts, over time and under commercial pressure, toward serving that group — whatever the founding principles say. So if you want to predict how ChatGPT or Perplexity or Google’s AI Mode will treat your citations in 2027, do not read the blog post announcing their principles. Read the invoice: who is being billed, and for what.

Apply it to the two protagonists. Perplexity, having killed ads, is funded by subscribers — the $20/month Pro tier and the $5/month Comet Plus add-on, plus usage-based enterprise credits. Its paying customer is a person who wants a trustworthy, well-cited synthesis and is willing to pay to keep advertising out of it. Leadership’s stated reason for going ad-free, per Stan Ventures, was that an answer engine people pay for has to stay objective. Notice that this is not altruism; it is customer-fit. The subscriber is buying trust, so the platform sells trust, so paying for good citations is consistent with the revenue model rather than a cost to be minimised.

OpenAI, on the free and Go tiers where ads run, is increasingly funded by advertisers. Its ad customer is a brand buying attention next to an answer. OpenAI has committed to “answer independence” — ads sit in labelled boxes and, by rule, do not influence what the model says. That commitment is credible precisely because it is aligned with OpenAI’s interest: the answer’s value collapses if users think it is for sale. But answer independence is a promise about the answer, not about the publisher. Nothing in the advertiser-funded model gives OpenAI a reason to route money to the outlet whose content was cited, because the advertiser does not care whether that outlet gets paid. Same principle, opposite result: the values track the customer, and the customers are different.

The mechanism is worth spelling out, because it is what makes the principle predictive rather than cynical. No one at either company has to be acting in bad faith for the drift to happen. A platform makes a thousand small product decisions a year — how prominently to show a citation, whether to add a click-through, how much of the source to summarise in place, whether to fund a publisher program or quietly let it lapse. Each decision is resolved, at the margin, in favour of the group whose money keeps the lights on, because that is what the incentives reward and what the metrics measure. Over a year those thousand small decisions compound into a posture. That is why you predict the posture from the payer, not from the founding memo: the memo is written once, the incentives operate every day.

This is why “trust-first” and “ad-funded” are not two points on a morality scale. They are two answers to a business question — where does the money enter? — and each answer carries an implied treatment of publishers that no press release will state plainly. The Revenue-Flow Ledger makes that implied treatment explicit.

The Revenue-Flow Ledger

The ledger traces four steps for any AI platform, then reads a fifth off them. Who pays the platform. What the platform sells that payer. Where your content sits in that transaction — paid supplier, unpaid fuel, licensed input, or traffic recipient. What the platform is therefore incentivised to do with your citation over time. And, as the summary column, whether the platform earns more when it sends value to you or when it keeps that value on-platform. The last column is the alignment score, and it is the one worth predicting behaviour from.

Run it across the three surfaces a UK publisher actually has to reckon with in 2026: ChatGPT’s ad tiers, Perplexity’s subscription model, and Google’s AI Mode. Read across each row before comparing down the columns.

The pattern is not subtle once the flow is laid out. Two of the three surfaces bill advertisers and owe you nothing when they cite you; one bills subscribers and pays you a share. The alignment column is not a judgement about which company is nicer. It is a prediction: the misaligned surfaces will, under commercial pressure, drift toward resolving more in-answer and leaking less to you, while the aligned surface has a structural reason — however thin the current cheque — to keep paying for the citations its subscribers value.

Why the ad-free platform is the one that pays you

The counter-intuitive result deserves to be stated flatly. The platform that positioned itself as trust-first and refused advertising is the one that puts money in publishers’ pockets. Perplexity’s Publishers’ Program, formalised as Comet Plus in January 2026, pays participating publishers 80% of subscription revenue against an initial $42.5M pool, per Tech Jacks Solutions and LLM Pulse; by Q1 2026 more than 2,400 publishers had signed on, including CNN, Condé Nast, Fortune, the Washington Post and Le Monde, per AI Business Weekly. When your content is cited in an answer a subscriber paid to see, a slice of that subscription flows to you.

OpenAI’s ad model does the reverse. In June 2026 the Press Gazette reported that OpenAI is not planning to share advertising revenue with publishers. OpenAI’s head of media partnerships, Varun Shetty, told publishers he did not see traffic as the “core value” of appearing in ChatGPT search — the value, in OpenAI’s telling, is the attribution and the citation itself, not a payment and not necessarily a click. OpenAI does pay some publishers, but through up-front licensing deals negotiated with a handful of large outlets, not an ongoing per-citation share open to all. As LLM Pulse put it, OpenAI pays up front for access while Perplexity pays on an ongoing basis when content is used. Those are opposite architectures, and they fall straight out of the ledger: the subscriber-funded platform pays its suppliers, the advertiser-funded platform pays its ad-sales pipeline.

A third path exists and is worth naming so the comparison is not a false binary. ProRata AI has said it will share 50% of ad revenue with the publishers whose content appears alongside its answers — an ad-funded model that does share back. So “ads” and “pays publishers” are not inherently opposed; OpenAI’s choice not to share is a choice, not a law of ad-funded search. That matters because it means the ledger is reading OpenAI’s specific incentives, not condemning advertising as such. If you are weighing how a paid campaign fits alongside earned coverage, the mechanics of buying links versus earning them are covered in our guide to sponsorship-based link building; the point here is narrower — who keeps the money the citation generates.

The catch: alignment is directional, not generous

If the article stopped there it would be cheerleading for Perplexity, and it should not, because the pro-publisher reading collapses under three facts that a fair analysis has to hold at the same time.

First, the purse is tiny. Eighty per cent of a $42.5M pool is about $34M; spread across 2,400 publishers that is roughly $14,000 each per year on average — and the split is usage-weighted, so a mid-sized outlet realistically sees low single-thousands. Against a real content operation’s costs, that is rounding-error money, not a business model. Second, Perplexity is being sued by the very publishers it pays: the New York Times, Dow Jones, the BBC and News Corp have all brought or backed copyright actions over how Perplexity gathers content, per Tech Jacks Solutions. “Pays publishers” and “accused of taking from publishers” are both true, simultaneously. Third, the aligned surface is small: Perplexity runs around 45 million monthly actives against ChatGPT’s 900 million-plus weekly users, per TechCrunch’s scale figures — so even a well-aligned share is a share of a far smaller pie.

There is a structural fragility on top of the small purse, and it sharpens rather than softens the ledger reading. Having killed advertising, Perplexity has left itself a single growth lever — subscription — and the publisher pool grows only if that lever pulls. Meanwhile its cost base is enormous and rising: a $750M three-year Azure commitment in January 2026 for GPU capacity, and a $400M cash-and-equity Snapchat integration, both per getPanto’s 2026 figures, against a valuation of roughly $20B on $450–500M of annualised revenue, per Stan Ventures. The publisher payments are therefore hostage to subscription growth in a way an advertiser-funded model’s revenue is not. That is not a reason to distrust the alignment — it is the alignment, made visible: the money that flows to publishers rises and falls with the money subscribers pay, exactly as the flow predicts, which is precisely why the direction is trustworthy even while the amount is not guaranteed.

And the misaligned side is not without value to publishers. OpenAI’s up-front licensing cheques, where they are offered, can dwarf a year of Perplexity rev-share for a large outlet; a national newspaper may rationally prefer one negotiated OpenAI or Google licence to thousands of thin per-citation micro-payments. The advertiser-funded surfaces also offer something the aligned one cannot: reach. A citation on ChatGPT or in Google’s AI Mode reaches an order of magnitude more people, and reach has brand value even when it neither pays nor clicks through. So the honest scorecard is that no model is cleanly good for publishers. The subscriber-funded one is aligned but poor; the advertiser-funded ones are rich and vast but keep the money. The ledger does not tell you one is virtuous. It tells you which direction each will drift — and that is still the more useful thing to know.

The Publisher’s Position Test

A ledger reading is only useful if it changes what you do. The Publisher’s Position Test converts each platform’s alignment score into one of four postures. The posture is chosen from the revenue-flow reading, never from the platform’s marketing — that is the whole discipline of the test.

• Revenue Partner. The platform pays you when cited and the payment clears a threshold worth the operational effort of enrolling and tracking. Posture: enrol, tag your cited URLs, and optimise citation volume partly for the share. This is where the aligned, subscriber-funded surface can land — if, and only if, your citation volume is high enough that low-thousands becomes worthwhile.
• Traffic Channel. The platform sends referral clicks that convert, even if it pays no share. Posture: optimise the citation for the click-through — the on-page promise, the accessible source page, the reason to leave the answer. AI referral traffic converts at roughly 14% versus about 3% for classic organic, per Criteo-aligned figures, so a click here is worth several ordinary ones.
• Brand Surface. The platform cites you to a vast audience but sends little traffic and no money — the answer resolves in place. Posture: optimise the citation for the mention itself. Get named, get framed favourably, build the branded and navigational demand that survives even when no one clicks. This is where the large advertiser-funded surfaces sit for most publishers.
• Threat. The platform harvests your content, returns neither money nor meaningful traffic, and competes with you for the user’s attention and intent. Posture: defensibility — licensing negotiation if you are large enough, owned distribution you control, and the access decisions covered in our work on negative-SEO and content defence and on the Google disavow tool in 2026. A Threat surface is not one to abandon — the citation may still carry brand value — but it is one to invest in on your own terms, not the platform’s.

The same platform can occupy different postures for different publishers. For a small B2B specialist whose citations are rarely enough to clear Perplexity’s payout threshold, Perplexity is a Traffic Channel, not a Revenue Partner. For a national newspaper with a licensing team, ChatGPT is a licensing negotiation rather than a pure Threat. The test is run per publisher and per market. If your market is outside the platforms’ core geographies, the postures shift again — the dynamics we cover for international link building and specifically for link building across India and South Asia apply, because citation display, licensing appetite and rev-share coverage all thin out away from the US and UK.

The invariant both bets leave standing

Step back from the divergence and something the two models share comes into focus, and it is the most important thing in the article. On both platforms, you cannot buy your way into the answer. OpenAI enforces this by rule — answer independence keeps ads out of what the model says. Perplexity enforces it by having no ads at all. So the one thing money cannot purchase on either surface is the citation itself. Whatever the ad question resolves to, the earned citation is the asset that survives it.

That reframes the whole Perplexity-versus-OpenAI contest. The business-model war is not a war over whether you can be cited. It is a war between the platforms over how to divide the money a citation creates — whether that money comes from advertisers or subscribers, and whether any of it flows back to the publisher. The citation is the invariant; the monetisation around it is the variable. Our earlier analysis of the earned-versus-paid split argued that the two are non-substitutable and must be sequenced rather than traded off; this article adds the layer beneath that — the platform’s business model determines whether your earned citation also comes with a share of the spoils, or merely with exposure. The scrutiny piece on answer independence made the related point that independence is a narrow technical guarantee, not a broad one; the same narrowness is why the citation stays un-buyable while everything around it is up for sale.

For a practitioner this is liberating, because it means the correct primary bet does not depend on predicting the winner of the ad wars. Earn the citation and it pays off under Perplexity’s model (as referral, brand and possibly rev-share), under OpenAI’s model (as referral and brand), and under Google’s (as brand and defensive presence). You are buying an asset that is priced into every scenario. The mechanics of earning it — the competitor citation and backlink analysis that shows you who is already named, and the guest posting and contributed-content work — and the niche-edit placements that get you into existing cited pages — that gets you into the corroboration are unchanged by which company wins the monetisation fight.

What actually pays: the rev-share is a signal, not a line item

It is tempting, having discovered that one platform pays publishers, to chase the payment. That would be a mistake, and seeing why sharpens the whole strategy. On the platform that does pay you, the payment is almost never the reason to optimise. Run the arithmetic: a mid-sized publisher’s realistic Comet Plus share is low single-thousands a year. The same publisher’s citations, if they drive referral traffic that converts at around 14% against 3% for organic, are worth far more in pipeline than the rev-share is in cash — often by an order of magnitude. The cheque is real, but it is a rounding error next to the citation’s downstream value.

So the rev-share’s true function is not revenue. It is information. Its presence tells you the platform’s incentives point toward keeping you supplied and cited — the alignment the ledger predicted, confirmed in cash. Its absence on the ad-funded surfaces tells you the opposite. You optimise for the citation’s referral and brand value on every platform; you treat the rev-share as a tiebreaker and an alignment signal, not as a line item to maximise. A publisher who reorganised their whole citation strategy around Perplexity’s payout would be optimising the smallest number on the page.

This is also why the media-buyer’s instinct — “put the budget where the ads are” — misfires for a publisher. The ad platforms are where you spend; the citation is where you earn, and it earns on all of them. The two decisions run on different budgets and different clocks, and conflating them is how content teams end up pouring effort into the surface that returns them the least. If you are deciding where a finite content and outreach budget goes, the link-building specialist’s allocation question is “where does an earned citation compound?” — not “where are the ads?”

A publisher’s response, worked

Make it concrete. Harbourline is an invented but specific case: a UK B2B publisher covering treasury and cash-management software, roughly £1.2M in annual revenue, earning citations across ChatGPT, Perplexity and Google’s AI Mode for queries like “best cash-flow forecasting tools for mid-market finance teams.” Its finance director wants a one-page answer to a blunt question: where do we invest to be cited, and how do we treat each platform? Run the ledger and the position test, with numbers.

Perplexity. Subscriber-funded, pays 80/20. Harbourline is cited on perhaps 300 relevant Perplexity answers a month. Enrolled in the Publishers’ Program, its realistic annual share is around £2,500 — real, but trivial against its cost base. However, Perplexity’s audience is high-intent researchers, and its referral converts well; those 300 monthly citations drive an estimated 40 site visits that convert at ~14%, worth far more in pipeline than the £2,500. Verdict: Revenue Partner on paper, but managed as a Traffic Channel — enrol for the share and the alignment signal, optimise for the click.

ChatGPT. Advertiser-funded, pays nothing per citation, but reaches a vast audience. Harbourline is cited on ChatGPT far more often — say 1,100 answers a month — yet most resolve in-answer with no click and no cash. Verdict: Brand Surface. The posture is to be named and framed well, and to convert that exposure into branded search demand that shows up later on channels Harbourline controls. No rev-share to chase; the value is the mention and the brand lift.

Google AI Mode. Advertiser-funded, keeps click and dollar on Google’s surface, licensing only for giants. For a £1.2M publisher there is no licence on offer. Verdict: Brand Surface shading toward Threat — optimise the citation for visibility, but do not depend on the traffic, and keep owned distribution (newsletter, direct) strong as a hedge against a surface built to retain the user.

The allocation that falls out: Harbourline puts the bulk of its earning effort into the citation itself — the studies, the contributed expertise, the featured-snippet-grade source pages that get it named — because that asset pays on all three surfaces. It enrols in Perplexity’s program (an afternoon’s setup for a small recurring cheque and a clean alignment read), treats ChatGPT and Google as brand surfaces to be won on framing rather than clicks, and builds branded demand as the moat against the two platforms that return it the least.

Now watch the naive reading misfire. A team that took the trade-post framing — “Perplexity dropped ads, so ignore it; put everything into the ad platforms” — would have walked past the only surface that pays it, optimised hardest for the surfaces that keep the money, and mistaken reach for return. Timeline: in month one both approaches look identical, because citations take time to compound. By month six, Harbourline’s ledger-driven allocation is drawing converting referral traffic from Perplexity, banking a small but real rev-share, and accruing branded demand from its ChatGPT presence — while the naive team is still buying ad clicks on the surface where it is most substitutable and wondering why its earned presence is not paying. The error was reading the ad decision as the strategy, when it was only ever a symptom of the revenue flow underneath.

When the argument is wrong

The strongest objection to this article is not that the ledger is inaccurate. It is that the ledger’s central claim — that Perplexity’s revenue model aligns it with publishers — is cosmetic. State it at full strength: “A $42.5M pool split across 2,400 publishers is marketing, not economics. Perplexity is being sued by the New York Times, Dow Jones and the BBC for scraping the same content it claims to pay for. You cannot call a platform ‘aligned with publishers’ while it fights them in court and pays them rounding-error sums. The 80/20 split is a press release with a dollar figure attached, and the whole revenue-flow thesis is naive for taking it at face value.” That is a serious objection, and every fact in it is true.

Concede all of it. The pool is trivial, the lawsuits are real, and the two coexist. But the thesis was never that Perplexity is generous — it is that its revenue model gives it a reason to pay publishers that OpenAI’s ad model structurally lacks, and that directional incentives predict the direction of drift even when today’s magnitude is trivial. The lawsuits and the payments are not a contradiction; they are the same fact seen from two sides. Perplexity needs publisher content badly enough to both take it and pay for it, because its subscriber wants cited answers. A platform that did not need the citation would do neither. OpenAI, whose ad customer does not care about the citation’s source, has no equivalent pull toward paying — which is exactly why it announced it will not.

So the right test is not “is $14,000 generous?” It is “does the money move in the direction the flow predicts?” — and it does. If Comet Plus subscriptions scale, the pool grows, because the incentive is real and load-bearing. If they stall, the whole model shrinks, payments included — which is the ledger predicting behaviour correctly, not a refutation of it. The magnitude is a function of subscription growth; the direction is a function of who pays. The article only ever claimed to predict direction, and the falsifying observation would be an advertiser-funded platform spontaneously deciding to share per-citation revenue with all publishers out of principle. OpenAI’s June 2026 statement that it will not do so is the ledger’s prediction confirmed, not contradicted.

The secondary objection — the media buyer’s — also survives contact only partly. “Perplexity removed itself from the ad buy, so for my next campaign the choice really is just ChatGPT versus Google.” True, and conceded: for a next-quarter media plan, buy where the ads are. But that horizon is not the publisher’s horizon. The ledger is a capital-allocation tool, not a campaign tool, and on the capital-allocation horizon it predicts that the misaligned surfaces will resolve more in-answer and leak less over time — which changes the future value of an ad click on them, and therefore informs even the media buy, just on a longer clock than the campaign runs on. The two objections share a shape: both mistake the near-term ad decision for the strategy. The whole point is that it is downstream of one.

The Monday test: what to do now

Five moves follow directly, and none of them requires knowing who wins the ad wars.

• Read the flow, not the release. For every AI surface you appear on, write the one-line ledger: who pays it, what it sells them, where your content sits, what it is therefore incentivised to do to your citation. Update it when the money changes, not when the marketing does.
• Enrol where rev-share exists — for the signal, not the sum. Join Perplexity’s Publishers’ Program if your citation volume clears a worthwhile threshold, but budget the effort as an afternoon and the return as a tiebreaker. Never reorganise a citation strategy around the smallest number on the page.
• Treat the vast ad surfaces as brand surfaces, and build branded demand. On ChatGPT and Google AI Mode, optimise for being named and framed well, then convert that exposure into branded and navigational demand that lands on channels you own. That demand is the moat against the platforms that return you the least.
• Negotiate or license if you are large enough; defend if you are not. The up-front licensing path is real but reserved for scale. Below it, defensibility is the play — owned distribution, and deliberate crawler-permission and disavow decisions so you are not feeding a Threat surface for free.
• Own the invariant. Put the bulk of your effort into earning the citation itself, because it is the one asset priced into every scenario — un-buyable on both bets, and valuable however the ad question resolves.

The publisher’s response to Perplexity’s ad-free bet and OpenAI’s ad model is not to pick the winner. It is to notice that both bets are fights over the money a citation creates, to read each platform’s incentives from its revenue flow rather than its principles, and to invest behind the one thing neither bet lets anyone buy. Do that and the ad wars become someone else’s problem — a fight over how to divide value you have already earned. For the broader playbook that this fits into, the core strategies hub, the what-is-link-building foundation, the current best-tools shortlist and the running 2026 link-building statistics carry the connected threads.