Can You Pay to Appear in ChatGPT Results?

The question splits in two the moment you look closely, and almost every argument on this topic comes from people arguing across the split without noticing. There is a paid surface in ChatGPT. There is also an unpaid surface. They are built by different teams, served by different systems, and crawled by different robots. Here is the split, stated plainly, before anything else.

Read the last row twice. It is the whole page in miniature. When a stranger emails offering to get your business "into ChatGPT," the honest version of that offer is either a media buy you could place yourself, or a body of editorial and technical work whose outcome nobody controls. There is no third product, however the invoice is worded.

What OpenAI actually started selling in 2026

OpenAI confirmed advertising in ChatGPT in January 2026 and began testing in the United States on 9 February 2026, later widening to Canada, Australia, New Zealand and the United Kingdom. The specifics matter more than the headline, because the specifics are what the cold emails blur.

Ads run for logged-in adult users on the Free and Go tiers. Paid tiers above that, Plus, Pro, Business, Enterprise and Education, do not carry them. Units are labelled as sponsored and set apart from the reply. OpenAI's own framing at launch, as reported by TechCrunch on the day, was direct: ads "do not influence the answers ChatGPT gives you," and conversations stay private from advertisers, who receive aggregate performance figures such as views and clicks rather than anything about individual users. The company also excluded under-18 accounts and sensitive categories including health, politics and mental health.

So the paid product exists and it is real. What it is not is a lever on the answer. An advertiser buys attention next to an answer. Nobody sells the answer.

On price, the honest position is that we do not have a reliable public figure. Secondary trade coverage since launch has reported launch CPMs near $60 with six-figure minimum commitments, later CPMs closer to $25, a cost-per-click option appearing in spring, and eventually self-serve access with no minimum at all. Those accounts do not reconcile with each other, and none of them is a first-party rate card. Treat any specific number you read, including the ones in the previous sentence, as unverified until you see it inside an advertiser account.

The architecture tell: OpenAI runs a separate robot for ads

If you want evidence that the two surfaces are genuinely separate systems rather than a marketing distinction, read OpenAI's public crawler documentation. Four bots are listed, each with a stated job.

Note what OAI-AdsBot does and does not do. It only ever sees a page because an advertiser submitted that page in an ad campaign. It performs a safety check. It is walled off from training. Nothing about that robot's job description gives it a route into ranking or retrieval, and OpenAI publishes it as a distinct user agent precisely so operators can treat it separately.

This distinction has a practical consequence most sites get backwards. Blocking GPTBot in your robots.txt is a defensible choice about training data, and it does not remove you from ChatGPT's search answers. Blocking OAI-SearchBot does. Sites that pasted a blanket AI-crawler block from a forum post in 2024 sometimes took themselves out of the exact surface they are now paying consultants to get them into. That one is worth checking today, and it costs nothing but a look at your own robots.txt file.

If you want the common configuration, where you decline to feed model training but stay eligible to be cited in answers, it looks like this. Visit yoursite.com/robots.txt and compare.

User-agent: GPTBot
Disallow: /

User-agent: OAI-SearchBot
Allow: /

What the fifty cold emails a day are actually selling

Business owners in Ontario are receiving a genuinely remarkable volume of outreach on this topic. The pitches vary in polish and converge in structure. Sorted by what is really on offer, they fall into five buckets, only one of which describes a real product sold under an accurate name.

1. "We will submit your business to ChatGPT"

There is no submission endpoint for answer inclusion. None exists in OpenAI's developer documentation, none is referenced in its help material, and the crawler documentation describes discovery by crawling rather than by registration. The advertising platform has a signup, and product feeds have an application process for merchants. Neither is a submission for citations. When this pitch is not simple invention, it usually resolves into someone setting up your Google Business Profile and calling that a ChatGPT submission.

2. "We have a partnership with OpenAI"

Commercial relationships with OpenAI are real: advertising accounts, merchant feed access, API contracts, enterprise agreements. Not one of them transfers influence over source selection, and OpenAI has publicly committed to the opposite. A vendor can truthfully say they hold an advertising account and then imply something far larger. Ask which agreement, ask what it entitles them to do, and ask them to point at the clause. The answer arrives quickly or it does not arrive.

3. "Guaranteed AI citations, guaranteed mentions"

This is the one that should end the conversation. The systems are closed, they change without notice, and the same prompt can return different sources on different days. A guarantee here is a promise about a machine the seller does not operate. We are unusually comfortable talking about guarantees, because we offer one, and ours is deliberately shaped to be keepable: if your existing domain shows no measurable organic search results after 12 months of work with Formative Digital, we work for free until you see them. That is a commitment about our labour, which we control. It is not a commitment about OpenAI's ranking, which nobody outside OpenAI controls.

4. "AI directory listing" and "AI index inclusion"

A newer variant sells placement in a directory the vendor operates, described as an index that AI systems read. Some of these directories are simply new websites with no authority and no meaningful audience. Paying to be listed on one is paying for a backlink from a page nothing cites. The tell is that a real citation source is one you had heard of before the pitch arrived.

5. Monitoring dashboards sold as placement

The last bucket has genuine value underneath a misleading label. Tools that track how assistants describe your brand, which competitors come up, and how often you are cited are useful instrumentation, and we use that category of tooling ourselves. The problem is the framing. A dashboard tells you where you stand. It does not move you. When a vendor charges a monthly fee whose only artifact is a monitoring report, you have bought a thermometer and been invoiced for a furnace.

Two of our research teardowns go further into how these offers are priced and worded: the line-by-line dissection of a $3,000 SEO package, and the catalogue of agency red flags that recur across the pitches owners forward to us.

The one place money touches the answer, and what it still does not buy

Honesty requires flagging an edge case, because it is the strongest counter-argument to this page and skipping it would be convenient rather than truthful. ChatGPT surfaces product results, and merchants can apply to send OpenAI a direct product feed. Money and a commercial relationship are involved. Does that constitute paid placement?

By OpenAI's own account, no. The company's shopping documentation states that product results are selected independently, are not ads, and are not influenced by OpenAI partnerships, with selection driven by signals such as price, availability, customer satisfaction, and whether the seller is the maker or primary seller of the item. A feed changes the freshness and reliability of the data the system holds about your catalogue. It does not change the weighting applied to it.

That distinction is worth internalising, because it generalises. Feeds, schema markup, a tidy Google Business Profile and consistent business details across the web are all mechanisms for making your facts legible and current. Legibility is not preference. You are removing reasons for a system to get you wrong, not buying reasons for it to prefer you.

And note the honest limit on all of this: OpenAI's statements about its own ranking are statements, not audited disclosures. No independent party can verify them from outside. We report them as the company's stated policy because that is what they are, and we would flag it immediately if observed behaviour started contradicting them.

Google leaned in, Perplexity walked out

ChatGPT is not the only surface, and the three major engines have taken visibly different positions on advertising, which tells you the question is live rather than settled.

Google has moved furthest toward monetising AI answers. Ads expanded into AI Overviews on desktop through 2025, reached eleven further English-speaking countries in December 2025, and arrived in AI Mode in early 2026. Perplexity went the other way. After testing sponsored follow-up questions and paid placements with named brands across 2024 and 2025, the company discontinued its advertising programme in February 2026, with an executive telling Search Engine Land that "we are in the accuracy business, and the business is giving the truth, the right answers," and arguing that sponsored placements erode trust in an answer engine even when clearly labelled.

The strategic reading matters for planning. Ad inventory in AI surfaces is expanding and will keep expanding, and it is a reasonable line in a marketing budget. The organic layer beneath it has stayed stubbornly unpurchaseable across three companies with very different commercial incentives, which is the best available evidence that it is likely to stay that way. That is an inference, not a fact, and we mark it as one.

What actually moves your odds

If the answer cannot be bought, the reasonable next question is what does work, and the research here is better than the marketing around it suggests. The foundational study is Aggarwal and colleagues at Princeton and Georgia Tech, published in November 2023 and benchmarked with GEO-bench across diverse query domains. They measured visibility improvements of up to 40 percent in generative engine responses from content-side changes alone, with the strongest gains coming from adding statistics, quotations and citations to authoritative sources. Keyword stuffing, the reflex tactic, did essentially nothing.

Google's own AI features documentation, which is first-party guidance rather than agency speculation, points the same way while explicitly deflating several popular tactics. It states that structured data is not required for generative AI search and there is no special markup to add, that content does not need breaking into small pieces for AI to understand it, that you should not write differently for generative systems, and that llms.txt files are ignored by Google Search. What it recommends instead is unglamorous: content people find genuinely useful, meeting the technical requirements to be crawlable and eligible for ordinary search.

This is Vector 1 of our twelve, Diagnose, more than anything else: find out what the assistants currently say about you before spending a dollar changing it. Our explanation of GEO sets out the full discipline and its research base if you want the ground underneath these five points.

On timelines, the shape of real progress is worth seeing once. In an anonymized engagement with an Ontario shipping container dealer, Google Search Console across a 16-month window, shared with permission, recorded 4,810 clicks against 627,000 impressions. Most of the first year sat at single-digit daily clicks. By early summer 2026 the same site was posting seventy-click days against six to nine thousand daily impressions. Nothing was bought. The curve simply took eleven months to bend, which is the part no cold email will ever tell you, and outcomes vary by market, competition and how much ground the site starts with.

Seven questions that sort the real from the rest

Run these against the next pitch that lands. They are ordered by how quickly they end a bad conversation, and none of them requires you to know anything technical.

A vendor who answers all seven comfortably may still be wrong for you on price or fit, but they are selling something real. Our longer treatment of the buying decision, including what to compare between providers, is in how to choose an SEO agency.

If you already signed with one of them

A fair number of owners reach this page after paying rather than before, which is an awkward position but not a hopeless one. The useful move is to find out what you actually bought, because in a meaningful share of cases the underlying work has some value and only the story around it was false.

Pull the last three invoices and the original proposal, and separate the promises from the deliverables. Then ask the provider for read access to whatever they have been doing: the pages they published, the changes they made to your site, the listings they claim to manage. If real artifacts exist, you may be looking at ordinary content and local search work that was mislabelled as AI placement to justify the price. That is salvageable, and often the honest conversation about renaming it goes better than expected.

If nothing can be produced, three steps are worth taking in short order. Confirm you control the assets: your domain registrar login, your website hosting, your Google Business Profile, your analytics. Reclaim anything held in an account you cannot access, because that bargaining position disappears once the relationship sours. Then read your termination clause before you send any email that signals you are leaving. Finally, if the pitch that sold you included a specific outcome promise, the Competition Bureau accepts complaints from the public, and the substantiation standard described below is the one that pitch would be measured against.

The Canadian angle: that email is a performance claim

What nobody outside these systems can tell you

A page like this earns its keep by being clear about its own limits, so here is the boundary of honest knowledge on this topic.

Nobody outside OpenAI knows the weighting inside its retrieval and ranking. The company describes its policy, and that description is the best evidence available, but it is unaudited and could change on a Tuesday without an announcement. Nobody can tell you why your competitor was cited and you were not on a specific prompt on a specific day, because the same question asked twice can return different sources. Nobody can promise that the current separation between advertising and organic answers survives commercial pressure indefinitely, and it would be foolish to assume permanence from three years of behaviour.

What can be said with confidence is narrower and still useful. No public mechanism currently exists to purchase inclusion in the organic answer on any major engine. The published documentation from OpenAI and Google describes discovery by crawling and ranking by content signals. The peer-reviewed work on this measures real, bounded gains from evidence-dense, well-structured content. Every one of those statements is checkable, which is more than can be said for the contents of your inbox.

Written by Matt Griffin, founder of Formative Digital, Brantford, Ontario. Published 2026-07-20.