ChatGPT Is Getting Ads. Here's How They Could Avoid Destroying Everyone's Trust

OpenAI announced they're bringing ads to ChatGPT. Cue the collective groan.

I get it. The knee-jerk reaction is to assume the worst, that ChatGPT will become another platform where the answer to "what's the best laptop?" is secretly sponsored by whoever paid the most. But it doesn't have to go that way. And from what I've seen, the industry is actually grappling with how to do this right.

I've spent the past few weeks digging into what's actually happening. Not the headlines, but the specific mechanics of how AI companies are experimenting with advertising. What I found was interesting: approaches that range from thoughtful to innovative to "this could go horribly wrong."

Why Chat Ads Hit Different

The core problem with advertising in AI assistants is that ChatGPT sounds like a trusted advisor.

When you ask Google a question, you know you're getting a list of links, some of which are paid. The transaction is transparent. But when you ask ChatGPT for advice, it responds in natural language, like a knowledgeable friend having a conversation with you. That changes everything. Academic research has confirmed what most of us intuitively feel: people struggle to detect embedded advertising in conversational AI. And when they do realise something was sponsored? They feel manipulated. Trust drops. Fast. Researchers call this the "fake friend dilemma", and it's the central challenge anyone monetising AI assistants needs to solve. The major players seem to understand this. OpenAI has been explicit that "ads do not influence the answers ChatGPT gives you." Perplexity has promised "the content of the answers will not be influenced by advertisers." Whether they can hold that line under commercial pressure remains to be seen. But at least the principle is stated.

The Approaches Being Tested Right Now

What struck me was the variety of models being explored. These aren't just banner ads shoved into a chat window. Some of these approaches are clever, and a few might actually make the user experience better rather than worse.

Ads Around Answers, Not In Them

The simplest approach, and probably the safest from a trust perspective, is to keep ads physically separate from the AI's response. OpenAI's stated plan is to place ads at the bottom of answers when they're relevant, clearly labelled and separated from the actual content. Google is already doing something similar with AI Overviews in search, showing sponsored results above or below the AI-generated summary. The answer itself stays untouched by commercial influence. The ad is just a module that happens to appear nearby.

Not particularly innovative, but that's partly the point. Users already understand this format from search. The familiar pattern buys trust.

Sponsored Follow-Up Questions Perplexity has pioneered an approach where ads take the form of suggested follow-up questions that are clearly marked as sponsored. The AI's answer remains independent, but below it you might see: "Sponsored: Want to compare prices on [Product Category]?"

The user has to actively choose to engage. You're not being sold to, you're being offered an option to explore something commercial if you want to. Even if you click a sponsored question, Perplexity claims the resulting answer is still generated by Perplexity, not written by the brand.

The risk? Sponsored questions could subtly shape what users become curious about. "Is Brand X the best?" is technically a question, but it's also doing persuasive work. This model needs tight editorial guidelines to avoid becoming manipulative.

The "Showroom" Model Microsoft is developing what they call "Showroom ads", immersive branded experiences that users can choose to enter. Think of it as the AI equivalent of walking into a branded store within a department store.

The assistant might say: "Want to explore the Mercedes-Benz showroom?" If you opt in, you enter a clearly branded environment where you can ask questions and get information directly from (or curated by) the brand. There's always a "return to neutral assistant" option visible.

I quite like this one. It treats advertising as a tool you can choose to use, not an intrusion you have to tolerate. The transparency is built into the structure. You know exactly when you're in a commercial context.

Outcome-Based Monetisation

This is the model I find most promising: don't charge for impressions or clicks, charge for completed transactions. OpenAI has launched "Instant Checkout" that works exactly this way. The product results you see are described as "organic and unsponsored". Merchants only pay a fee when you actually complete a purchase. And OpenAI explicitly states that enabling checkout doesn't influence how products are ranked.

This aligns incentives properly. The platform makes money when the user succeeds in buying something they actually wanted, not when they click on something shiny. It removes the temptation to prioritise advertisers over users because there's no advertising in the traditional sense. For anyone working in paid media , this shift from CPC to CPA models is worth watching.

The downside? It requires deep commerce integration and only works for products that can be bought directly. But for shopping-related queries, it might be the gold standard.

Where This Could All Go Wrong

I've outlined the optimistic scenarios. Now for the landmines.

The biggest risk isn't that ads exist. It's that ads get embedded into the assistant's actual voice. If ChatGPT starts casually mentioning products as part of its "helpful" response, with no clear labelling, trust will evaporate almost immediately. Research shows that once users realise they've been subjected to covert advertising, they rate the experience as manipulative and intrusive. Meta has already indicated they'll use interactions with AI features to personalise ads and content elsewhere. That's a very different approach, essentially treating your private conversations with an AI as advertising data. You can see how that might trigger a "you're mining my private thoughts" backlash.

The other concern is sensitive topics. Imagine asking an AI assistant about mental health and being served an ad for a therapy app. Or asking about financial difficulties and seeing sponsored content for a loan product. Google has explicitly excluded sensitive verticals like healthcare, finance and politics from ads in AI Overviews. A sensible firebreak that others would be wise to adopt.

What The Regulators Say

Some good news for those worried about a race to the bottom: regulators are already paying attention, and the rules are pretty clear. The UK's ASA requires that marketing communications be "obviously identifiable as such." The US FTC says disclosures must be "clear and conspicuous" and in interactive media should be "unavoidable." The EU's Digital Services Act goes further, requiring that ads are clearly labelled, show who paid for them, and explain why you're seeing them.

The regulatory framework already prohibits the worst abuses. The "trust-preserving" design path is also the "compliance-friendly" path. Companies that try to blur the line

between advice and advertising aren't just risking user trust, they're risking enforcement action.

A Constitution for AI Advertising

If I were advising an AI company on how to introduce advertising without destroying their users' trust, I'd suggest publishing, and actually enforcing, a small set of non-negotiable principles:

Answer independence: Ads never influence what the AI recommends or says. Ever.

Visible separation: The word "Sponsored" is always visible without effort. No hiding it in small print.

User control: Users can opt out of personalised advertising, and there's a paid ad-free tier for those who want it.

Sensitive topic exclusion: No advertising on healthcare queries, mental health discussions, financial distress, or political topics.

Privacy boundary: Conversation data is never shared with or sold to advertisers.

Then design ad formats that physically cannot violate these principles. OpenAI and Perplexity have both articulated versions of these commitments. Whether they hold them under commercial pressure will determine whether AI advertising becomes a sustainable business model or a cautionary tale.

So Where Does This Leave Us?

Ads in ChatGPT aren't inherently evil. Done well, they could fund free access to powerful AI tools while respecting users' intelligence and trust. Done badly, they could poison the well for the entire industry.

Everything I've read comes down to one principle: monetise the moment of intent without contaminating the advice loop. Ads can exist around answers, as optional paths, as transaction fees, as clearly marked commercial spaces. But they can never be allowed to pretend they're not ads.

The companies that understand this will build sustainable businesses. The ones that get greedy will watch their users leave the moment a more trustworthy alternative appears. And with AI reshaping how search and discovery work, getting this right matters more than ever.

Make money like a tool, not like a trick.