The industry took it as settled. Search has ads. Social has ads. Free apps have ads. So an AI answer engine, once it had enough users, would obviously get ads too. The only questions were when and what format. Perplexity even seemed to confirm the script: in November 2024 it became the first major AI search engine to sell advertising, and the trade press treated the move as the starting gun for a whole new ad channel.
Then, over the following year, Perplexity quietly took the gun back. It stopped signing new advertisers, let the program wind down, and by early 2026 executives were saying plainly that they had no plans to return to ads. The reason they gave was not weak demand or a bad quarter. It was something more awkward for the whole category: a user who sees sponsored answers starts to doubt every answer, and an answer engine that is not trusted has nothing to sell.
That is the tension worth understanding. Not whether Perplexity made the right call, but what its retreat exposes about a business model the industry assumed was inevitable. Ads and a trusted answer may not sit together the way ads and a list of blue links always did.
Origin: the first answer engine to put a price on the answer
To see why the retreat matters, start with how deliberately Perplexity entered advertising in the first place. This was not a reluctant bolt-on. The company published a blog post, Why we're experimenting with advertising, framing ads as a way to build a business that did not depend on subscriptions alone, and arguing that a healthy ad model could also send money back to the publishers whose work the engine drew on.
The format was designed to look careful. Perplexity did not sell the answer itself. It sold a "sponsored follow-up question" that sat alongside a response, a prompt such as a job-search question paid for by a recruitment brand. When a user clicked it, Perplexity's own model wrote the answer, and the brand could not write or edit it. Pricing ran on a CPM model, a fee for every thousand views, rather than pay-per-click. The pitch to advertisers and to users was the same: the sponsorship buys placement and a question, never the substance of the response.
The launch roster was credible. Indeed, Whole Foods, Universal McCann and PMG were among the first partners when sponsored placements went live for US users in November 2024. For a company that had spent its short life arguing it was a cleaner alternative to Google, this was the moment it accepted the same basic deal Google had run for two decades: give the answer away, charge someone else for proximity to it.
The design was thoughtful. The economics, it turned out, were tiny. Reporting later put Perplexity's advertising revenue for 2024 at roughly 20,000 dollars against total revenue of about 34 million, a rounding error. That gap matters for what came next, because it meant the trust question and the revenue question were never really in balance. There was barely any revenue to defend.
Present: why a launched product got switched off
Through 2025 the program did not grow into the channel the launch implied. It shrank. Following it in order makes the retreat legible.
In August 2025, Taz Patel, the executive hired to build Perplexity's advertising and shopping business, left after about nine months. Trade reporting noted that brand clients had given his work lukewarm feedback. In October 2025, at Advertising Week in New York, Perplexity's head of publisher partnerships said the company had stopped taking new advertisers and was reassessing how ads should fit the product, adding that ads would not come to its Comet browser for now. Then in February 2026, executives confirmed to the Financial Times that the exit was effectively permanent: no plans to pursue advertising further.
Two explanations sit behind that timeline, and the honest reading needs both.
The first is plain commercial weakness. Perplexity's ad product did not give buyers what performance advertisers expect. Marketers were interested but frustrated by limited availability, thin lower-funnel tooling and weak measurement, with no reliable way to track click-through or return on spend. With a user base far smaller than Google's or ChatGPT's, Perplexity could not compete for budgets on scale, and buyers raised brand-safety and CPM concerns. On those facts alone, a struggling product got cut. That happens constantly and would not be worth a blog post.
The second explanation is the one that travels. Perplexity's leaders did not only say the product underperformed. They argued the model itself was wrong for an answer engine. The line that summed it up, from a Perplexity executive to the Financial Times, was that with ads a user would just start doubting everything, which is why the company did not see ads as worth pursuing. The fuller logic: a person uses an answer engine because they believe it is giving them the best available answer, and they will pay for that belief; introduce a commercial interest into the surface and the belief is contaminated, not just on the sponsored answer but on all of them.
That is the part that should hold a marketer's attention. A blue-links search page can carry ads without the user doubting the organic results below, because the user has been trained for twenty years to see two zones on one page. An answer engine offers one zone. There is a single response, presented as the synthesis of everything the model found. There is no visible "organic" area for the user to retreat to. So a sponsored element does not sit beside the trusted thing. As far as the user's confidence goes, it sits inside it. Perplexity's bet is that this difference is structural, not a labelling problem to be fixed with a better "sponsored" tag.
The consumer data gives the argument some support without settling it. Polling cited by eMarketer found that 63 percent of US adults said ads in AI search results made them trust those results less. That is a survey of stated attitude, not measured behaviour, and people often tolerate in practice what they dislike in a poll. But for a product whose entire value rests on perceived neutrality, a clear majority saying ads erode trust is not a number to wave away.
The alternative: where Perplexity decided the money is
Walking away from ads only works if there is somewhere else to stand. Perplexity's case is that there are three other places, and that an answer engine is better suited to all of them.
The first is subscriptions, and it is the foundation. Perplexity sells Pro and Max tiers, and most of its revenue now comes from subscriptions rather than ads. The scale of that business is what makes the ad retreat affordable: the company raised funding in 2025 at a valuation reported around 20 billion dollars, and its annualised revenue climbed into the hundreds of millions over 2025 on subscription growth. When users are paying directly for the answer, the incentive is to make the answer good. When advertisers pay, the incentive splits. That alignment is the whole argument for the subscription model, and it is the cleanest version of "the customer is the user, not the advertiser."
The second is publisher revenue sharing. Rather than use ads to funnel money to publishers, Perplexity built Comet Plus, a subscription program that pays publishers out of a dedicated pool when their content is used through direct visits, crawler access or AI agent traffic. It opened the pool at 42.5 million dollars and keeps 20 percent, passing 80 percent to participating publishers. Whether the sums are large enough to matter to publishers is a fair and open question. But the design choice is consistent: fund the supply side from subscription money, not from ad inventory wedged into the answer.
The third is commerce, and it is the most revealing. Just before Black Friday 2025, Perplexity launched Buy with Pro, in-chat shopping built with PayPal, where a user researching a product can complete a one-click purchase through PayPal authentication without leaving the conversation. This is the subtle point. Commerce is still monetisation. Perplexity has not taken a vow against making money from a buying intent. It has chosen a model where it earns when the user actually buys something they wanted, which it argues keeps the engine pointed at the user's goal, rather than a model where it earns when a brand pays to be mentioned regardless of fit. Reasonable people can debate whether that line truly holds, especially as commerce recommendations carry their own bias risk. The commerce path has also already drawn friction: Amazon sent Perplexity a legal threat over its Comet browser agent making purchases on Amazon, which Perplexity publicly called bullying. Agentic shopping is not a settled business model. But it tells you Perplexity's objection was specific. It walked away from advertising, not from revenue.
Future and impact: can ads and a trusted answer coexist?
Set Perplexity's choice against what its largest rivals are doing, because the contrast is where the real lesson sits.
OpenAI moved the other way. It began testing ads in ChatGPT in early 2026, for logged-in adults on the free and Go tiers, with paying Plus, Pro and Enterprise users left ad-free. OpenAI's stated guardrails are precisely the ones Perplexity decided were not enough: ads are clearly labelled as sponsored, kept visually separate from the answer, and, OpenAI says, do not influence the content of the response. Google is pushing in the same direction from the other side, building Gemini-powered ad formats into AI Mode, including conversational ad units and shopping ads with AI-written explainers, all marked sponsored. Anthropic, meanwhile, has lined up with Perplexity, committing to keep Claude free of advertising and treating that as a selling point.
So the same question, can a commercial placement live next to a trusted machine answer, has produced four answers from four serious companies. That split is the finding. The market does not actually know yet.
There are two ways the next two years resolve it, and a marketer should plan for both.
In one outcome, OpenAI and Google make labelled ads in answers work. Users get used to a sponsored unit beside an AI response the way they got used to it beside a search result, the trust hit stays small, and ads become a normal layer of the answer-engine economy. If that happens, Perplexity will look like a company that was too cautious, or simply too small to make the ad model pay, and dressed a commercial failure in a principle. The pull of ad revenue at the scale of ChatGPT is enormous, and history leans toward free, ad-supported tiers winning the mass market.
In the other outcome, Perplexity is early rather than wrong. Doubt compounds quietly. Users who cannot tell how much a sponsored interest shaped an answer trust the engine a little less, ask it fewer of the high-stakes questions, and the answer engines that stayed clean become the ones people rely on for decisions that matter. In that world the durable model is subscriptions and commerce, and advertising never becomes the centre of gravity for AI answers that it became for search.
For anyone building a marketing strategy around AI search, the practical reading is this. Treat the ad-supported answer engine as a real but unproven channel, not a settled one, and watch the trust signal as closely as the reach numbers. Build presence in the engines that do not run ads as well, through being the source a model cites, because a meaningful share of AI search may stay ad-free by deliberate design. And notice the deeper shift Perplexity's retreat points at: in an answer-engine world, the asset is the user's belief that the answer is honest. That asset is harder to rent out than a slot on a results page, and the industry's old certainty that everything eventually carries ads is, for the first time, genuinely in question.
Council summary
This post argues that Perplexity's 2026 exit from advertising is not a one-company misstep but a signal that ads and a single trusted machine answer may be structurally hard to combine, because an answer engine offers no separate "organic" zone for a doubting user to retreat to. Every figure was web-checked against primary reporting: the November 2024 launch with Indeed, Whole Foods, Universal McCann and PMG; the roughly 20,000 dollars in 2024 ad revenue against about 34 million total; the 63 percent Ipsos finding cited by eMarketer; the 20 billion dollar valuation; the 42.5 million dollar Comet Plus pool with its 80/20 split; and OpenAI's February 2026 ad test. All checked out, so the edits stayed limited to style and pacing. The honest core is that four serious companies have landed on four different answers, so the model is genuinely unsettled. The takeaway: treat ad-supported AI search as a real but unproven channel, track the trust signal alongside reach, and build presence in the engines that stay ad-free by design.
Comments