Running Your First Paid Placement Inside ChatGPT and Perplexity Responses Step by Step

A marketer setting out to buy a placement inside a Perplexity answer in 2026 will not find one. Perplexity walked away from advertising in February 2026, moved to a subscription-only model, shut down its ad product entirely, and left behind no self-serve platform, no waitlist, and no managed-buy alternative. Any plan built on the assumption that Perplexity sells ad inventory needs to be scrapped before a single dollar moves, because the company's exit from advertising was a deliberate strategic choice.

Perplexity's own account of the decision, relayed to the Financial Times, centered on trust: once users noticed clearly labeled sponsored placements, they started doubting everything, questioning the integrity of answers that had nothing to do with the ad itself. For a product whose entire value proposition rests on the reliability of its answers, that doubt was an existential risk, and Perplexity chose subscription revenue over advertiser revenue as a lasting structural bet.

None of this makes Perplexity irrelevant to a marketing team. It makes Perplexity an earned-visibility surface rather than a paid one, and that distinction governs every tactic a brand can use there: showing up in a Perplexity answer today depends on being cited as a source, not on winning an auction. Reaching Perplexity's users means organic citation strategy, commonly called generative engine optimization (GEO), built around the kind of content and structured data that gets pulled into an answer, not a media buy negotiated with an ad sales team.

That reframing sets the real scope of this walkthrough. Paid placement inside an AI assistant's response means ChatGPT today, with Google AI Mode and AI Overviews reachable through existing Google Ads campaigns and Microsoft Copilot reachable through existing Microsoft Advertising accounts with automatic eligibility and no separate Copilot Ads Manager. Perplexity has no broadly available self-serve ad manager, and the program it ran before exiting advertising admitted fewer than 0.5% of applicant brands, a figure that signals how narrow that door already was before it closed. Claude sells no ad placement. What follows is a guide to the one channel where a marketer can actually open an account, set a bid, and launch a campaign inside a conversational AI product this week, and that is a narrower but far more actionable target than the premise of "advertising inside Perplexity and ChatGPT" first suggested.

How a ChatGPT ad works, format, placement, and audience constraints

A ChatGPT ad starts as a labeled sponsored card that sits below the model's answer, and it is never folded into the answer text itself. OpenAI has since added other formats, including a contextual sidebar placement and a format called Sponsored Agents, but the original and still-central unit is that bottom-of-answer card, clearly marked as sponsored content distinct from what the model generated.

That separation is not a cosmetic choice. OpenAI states explicitly that paid placements remain separate from ChatGPT's answers and that no advertiser can influence what the model actually says in response to a user's question. Placements that sit alongside a response hold up better over time than placements that try to shape the response itself, because the moment an ad starts coloring the answer, the user has reason to distrust the answer. Platforms built specifically for this channel are designed around that same boundary, reading the context of a conversation to decide where an ad belongs without touching the content of the reply the user receives.

The unit itself is small by design. A headline runs to a tight character limit, the description is capped at a short phrase, the image is square and bound by minimum and recommended sizes, and a brand favicon rides alongside the copy. Because the card renders compact and inline rather than as a large display unit, the discipline of writing tight copy carries more weight here than it does in a typical display campaign, a point the creative section returns to in more detail.

The audience constraint matters more than the unit's size. ChatGPT ads show only to logged-in adults on the Free and Go tiers, where CPM pricing ranges from $25 to $60, while Plus, Pro, Business, and Enterprise subscribers see no ads. ChatGPT's total user base is large, but the addressable ad inventory sits on a narrower slice of it, skewed toward consumer and casual use rather than the paid tiers where many professional and business users do their work. That is a planning input: a B2B team targeting senior decision-makers should know going in that the people most likely to be on a paid, ad-free tier are exactly the buyers it may be trying to reach, and should size expectations and budget around the Free and Go audience that the ad product actually touches.

Targeting logic: context hints instead of keywords

ChatGPT advertising replaces the keyword with something OpenAI calls a context hint, and learning to write one well is the highest-leverage skill a marketer can bring to this channel. A context hint is a freeform, natural-language description of the conversations where an ad belongs, set at the ad group level and capped at 280 characters. It is not a list of keywords, not a set of demographic selectors, and not a remarketing audience built from past site visits. The system reads the hint against the live conversation in progress, so it matches on conversational context instead of parsing a query string the way a search engine parses a typed phrase.

Placement is decided by what OpenAI's own documentation describes as a relevance-weighted, second-price auction. If it's second-price, an advertiser pays just above the next competing bid, not the full amount of its own maximum bid. Relevance-weighted means a more precisely targeted ad can beat a higher raw bid if its context hint, copy, and landing page line up more tightly with the conversation at hand. A well-written hint from a smaller advertiser can outcompete a vague hint from a larger one, an unusual property in paid media that rewards craft over budget size, especially while most advertisers in the channel are still new to writing for it.

The practical failure mode is predictable: teams that write hints as comma-separated topic lists, the habit search advertising trained into a generation of marketers, get weak matching. A hint should describe what the person asking is trying to accomplish and why the advertiser's product is genuinely useful to them at that moment, not simply name the product category. OpenAI's own auction documentation confirms that the system weighs the context and intent of the current conversation alongside the ad's landing page, title, and copy, so creative quality and landing page relevance feed directly into relevance ranking.

What makes this worth the effort is the nature of the intent being expressed. A user typing a question inside a ChatGPT conversation has already stated a specific need in full sentences, often with the constraints and the stage of their decision built into the phrasing, which hands an advertiser more signal than a two- or three-word keyword ever carried. That is the core idea behind real-time prompt analysis in conversational AI advertising: the conversation itself is the signal, and the systems built to read that signal as it happens can match ads with a precision that keyword-matching was never built to reach.

Account setup and campaign structure in OpenAI Ads Manager

You start getting a campaign live by getting access, and access has gotten a lot easier. The self-serve Ads Manager is now open to any eligible advertiser, and the large spend commitment that gated the platform at launch has been removed. In April 2026, OpenAI dropped the minimum spend significantly, to $50,000, and brought CPC bidding into a limited pilot alongside the existing CPM model, with full CPC availability arriving on May 5, 2026, when OpenAI's own announcement confirmed it was "adding cost-per-click (CPC) bidding" to the self-serve manager.

The first decision inside the manager is objective selection, and it determines the bidding model the rest of the campaign runs on. A Views campaign optimizes for visibility and is priced on CPM. A Clicks campaign optimizes for traffic and is priced on CPC. OpenAI publishes current CPM and CPC bid ranges in its Ads Manager guidance, and a marketer should check that documentation at setup time rather than anchor to figures that may have shifted since.

Campaign structure above the ad group is a campaign that holds a budget and an objective, with ad groups sitting underneath it. The ad group is where this format departs from convention, because each ad group carries its own context hint rather than a shared set of keywords. So you should organize ad groups around territories of intent, the kinds of conversations a product is genuinely useful inside, rather than along product-line or item-category boundaries the way a search account often is.

Geographic reach has also expanded since launch. Ads initially targeted only logged-in adults in one country, and availability has since grown to more than 40 countries, with reach extending to logged-out users as well. Because this expansion has continued past publication, a marketer setting up a first campaign should confirm current geographic availability directly inside the platform.

The sequence in practice runs in a consistent order: pick the objective, set the budget, build the ad group and write its context hint, then upload the creative. Each step constrains the one after it. Objective selection, the first click in the manager, deserves more thought than it usually gets from a team used to defaulting to whatever bid type feels most familiar from search.

Bidding strategy for a first campaign

The choice between CPM and CPC should follow directly from what the campaign is trying to prove, not from habit. A Views campaign, priced on CPM, fits a goal built around brand awareness or category presence, where the advertiser is paying for exposure regardless of whether any individual user clicks through. A Clicks campaign, priced on CPC, fits a goal built around measurable traffic and downstream conversion, where the advertiser pays only when a user actually follows the ad to a landing page. CPM pricing spans a moderate range, and the default max bid at pilot launch sat at the top of that range, so you should know this before setting a first bid, or the starting number will look artificially low by comparison.

Recommended CPC bids fall in the low single digits. Because the auction is second-price, the max bid a marketer sets functions as a ceiling rather than the price actually paid, which means setting a max bid meaningfully above the suggested range costs nothing extra unless a competing advertiser forces the price up to meet it. Underbidding carries the opposite risk: a bid set too low simply loses placements. The minimum daily budget required to run a campaign is low enough that a test campaign can generate a real read on performance within a few weeks without a large financial commitment, which argues for treating the first few weeks explicitly as a learning phase and holding off on bid optimization until that signal has had time to accumulate.

What makes copy work inside a conversational answer

Most marketers running their first campaign in this channel already have well-developed instincts for display and search creative, instincts that do not transfer cleanly to a unit sitting directly below a natural-language answer. Conversational placement punishes generic promotional copy more severely than display or even search does, because the user reads a genuinely useful AI answer and a vague, templated ad right beneath it in the same breath, and the contrast is immediately visible.

The unit's small, inline rendering means tight copy carries more of the persuasive weight than image art does, and every character in the headline has to earn its place given the tight limit it runs against. Because the ad sits immediately below an answer that just addressed a specific question in the user's own words, a generic tagline reads as especially out of place in that moment, and copy that matches the register of the conversation the user was just having will outperform copy written for a broader, untargeted audience.

The auction reinforces this directly: it reads the context hint and the ad copy together, so a precise hint paired with copy that answers the same underlying intent will beat a broad hint paired with clever copy that doesn't actually match it. Landing page relevance feeds the same auction. If a landing page resolves the specific intent the hint describes, it supports eligibility, but a generic homepage does not carry the same weight no matter how strong the ad copy above it is.

The simplest way to frame the creative job is as an extension of the conversation. An ad that reads like the next sentence a knowledgeable person would offer, given what the user just asked, will perform better than one that pivots abruptly into sales language the moment the answer ends.

What attribution cannot yet tell you

The measurement infrastructure behind ChatGPT ads is still new, and OpenAI itself states the platform does not yet have performance benchmarks across advertisers, industries, or campaign types. So you should read figures reported from any individual test, including the ones below, as directional, not as an industry norm.

A documented case from B2B testing illustrates the gap: one campaign, tagged with tracking parameters, showed OpenAI's platform reporting 57 clicks against fewer than 20 visits recorded in a web analytics tool, and a separate test reported 53 platform clicks against 35 sessions recorded in that same tool. That gap reflects a structural feature of how discovery works inside an AI conversation: a user can see an ad, not click it, keep researching inside the same conversation thread, and convert later through an entirely different path that no tracking parameter ever captured. A marketer applying a last-click attribution model to that behavior will undercount the channel's actual influence on the outcome.

A first campaign should still track a defined set of numbers, even with that gap acknowledged. Impressions and CPM or CPC actuals against the targets set at launch confirm that the auction is functioning and that the context hint is matching conversations as intended. Click volume and landing page behavior show you whether the copy and the landing page are actually converting the intent the hint promised to deliver. If branded search lift rises across the campaign period, that offers a downstream signal that the channel is building awareness even when it never produced a click. Assisted conversions, pulled from any attribution model that looks further back than the last touch, help recover some of the influence that last-click reporting misses by design.

None of this is a reason to withhold budget from the channel. It is a reason to set expectations correctly before the campaign launches. The data gathered during a first campaign, the hint's match rate, the copy's click-through, the branded search movement, becomes the baseline against which a second campaign gets built, bid, and targeted with far more confidence than the first one could ever have.