Redefining the relationship between AI utility and digital advertising.

AI usage disclaimer: AI tools were used for editing visual assets, writing feedback, assistance in locating relevant sources, & Chicago-style citation formatting; all early drafts, ideas, arguments, & experiences in this article are my own.

Taxing advertising

I have a good friend & mentor at work, and we’ve been exploring new AI trends in tech and design. He brought up the news that OpenAI recently announced that it would begin testing ads in the free and Go tiers of ChatGPT in the U.S., with the promises that ads will be clearly labeled & won’t influence answers.¹ ² We’ve both fought against needless promotional content before and lamented that frontier AI platforms are falling into the same pattern. As designers and users, we’ve learned that “free” usually means putting up with interruptive, slightly creepy ads that feel more like a tax than a benefit — a frustration tax that now colors how we approach free‑tier services and now AI tools. However, we started chatting through it a bit more, and explored whether there might actually be some good that can come out of it.

While advertisements in free services like ChatGPT risk frustrating users & harming brand trust, frontier AI models have a unique opportunity to replace the traditional “Sponsored” tag with promotional features that genuinely enhance the utility of threads & responses instead of undermining them. Before exploring how digital ads might actually do this, I want to briefly revisit how ads evolved on TV and YouTube, then look at the unique opportunities frontier AI platforms have, and finally ground it in a simple electrician problem to show how ads could enhance an AI chat thread.

New platform, same commercial

Billboards, cable TV commercials, and radio ads typically felt like a jarring distraction — pulling me out of driving, the show I was watching, or the song I was listening to. Further, the commercials themselves were usually irrelevant to whatever I was interested in. But since there was less focused demographic targeting on TV, aside from specific channels that entertained a certain subject like Comedy Central, Nickelodeon, TV Land, etc., it made sense that you would see a wide variety of advertisements ranging from uplifting prescription drug commercials to wacky toy ads for kids. And I remember when I started seeing that same pattern begin on YouTube videos and feeling frustrated by that since it just felt like TV commercials were coming to this new platform.

YouTube used to feel like a ‘free expression’ platform for anyone & everyone to make videos on the internet. It gave people the means to show off talented, creative, weird, strange, hilarious, and odd content that we used to expect from YouTube and the TV shows that would curate its content like Tosh.0 or Ridiculousness. Even though ads came onto the scene just a year later, YouTube’s CEO Chad Hurley put his ad approach this way:

We think there are better ways for people to engage with brands than forcing them to watch a commercial before seeing content. You could ask anyone on the net if they enjoy that experience and they’d probably say no.
 — Chad Hurley ³

But when YouTube was acquired by Google, ads started to function like commercials on TV by the end of 2007, and at first they were annoying, but at least they were short, skippable, and rarely shown.⁴ Today, however, ads can last over 90 seconds if the user allows it, run multiple ads in basically what is a de facto commercial break, and not all ads offer the option to skip until the timer cools down.

But YouTube ads felt more targeted, which was both helpful and creepy at times. If I started watching skateboarding videos, I might see advertisements for things skaters are interested in (e.g., skate decks, certain clothing brands, X Games announcements, and so on). With data sharing, I’d start seeing ads based on things I searched for online or just after I made an online purchase for a similar item — the latter of which always felt odd to me since I already bought it. Still, it was better in a sense and more dialed in to relevant audiences than traditional TV or radio commercials.

All of this has taught us to expect that “free‑tier” experience means interruptive, slightly creepy, and rarely on our terms. That’s the mental model users will bring to free‑tier plans with ads in AI tools, and it functions like a frustration tax: “free” feels less like a gift and more like a tax you pay in attention. The opportunity for frontier AI is to break that pattern by making promotional content feel like a feature that even paid‑tier accounts may want to keep, so paid plans can focus on adding capabilities rather than simply removing frustration taxes.

Strengths of the AI pioneer

Shifting over to ChatGPT and other frontier AI platforms, the obvious risk is that ads will simply adopt the same look and feel they have on search engines — those “sponsored” results at the top of the page — but these tools are capable of much more than that. While I don’t have the exact data in front of me, I often wonder how frequently people scroll past those sponsored results to reach the first “organic” link and how often accidental clicks leave them feeling misled instead of helped. One recent roundup of digital ad research found that around 86% of users report banner blindness (effectively ignoring banner‑like advertising) and that average banner engagement rates hover around 0.06%.⁵

If frontier AI platforms adopt a similar pattern, I imagine the analytics data won’t change much. However, these platforms aren’t doing the same things as search engines are either — they do a lot more, obviously. They’re great at synthesizing content from a variety of sources based on the content of the user’s prompt. And they’re quickly doing more and more. For instance, I’m continuously impressed with Perplexity’s updates and new features it adds.

Advertisements can also be better with AI platforms. If search engines, YouTube, and other digital platforms can have more relevant ads just based on the search query, content of the video, or knowing the target audience, they can do all the more with a user’s prompt & chat thread. Even a basic prompt contains far more data than a general search query, and users are far more likely to write in natural language to AI platforms than to traditional search boxes. This can include nuanced details, context, and specificity that wasn’t as easy to do with a search engine — even with advanced search tools. Further, users no longer have to necessarily scan through each webpage result in a response to figure out what’s relevant to what they’re looking for. Whether or not they ought to is another question, but the sources in the response are very likely to be on topic with the prompt anyway.

Discovering marketed electricity

A month or so ago, several outlets in our kitchen stopped working after a GFCI outlet tripped, and a quick do-it-yourself (DIY) attempt led to a small spark behind the outlet. At that point, the problem moved from “easy DIY project” to “I need a professional who won’t overcharge me or cut corners. So imagine I turn to an AI assistant and type something like:

I have three outlets that went out in our kitchen. I replaced a GFCI outlet, but when I pulled it back out to check the wiring I saw a small spark behind the outlet. I’m comfortable with basic DIY, but I don’t want to risk an electrical fire or make the problem worse. I’m looking for an electrician near me who is thorough, is transparent about pricing, and won’t pressure me into unnecessary work. Give me a few options and explain which one you’d choose for this situation.

This is a perfect example of the kind of urgent, high‑anxiety scenarios where ads could either quietly undermine trust — or actually help. In the rest of this section, I’m treating my electrical experience as a small test case for this idea: what happens if we replace a thin “sponsored” tag with richer promotional features like coupons, adjacent follow-up content, even a contextual panel that doesn’t distract from the main thread so much?

So, how might ads be smartly introduced from a prompt like this? Well, maybe there are a few ways. First, imagine a fairly conventional AI response: a ranked list of three electricians with one of them labeled “Sponsored.”

This is the default pattern we’ve inherited from search: a solid comparison list with a tiny “Sponsored” tag bolted onto one option. It technically discloses the ad, but it also quietly introduces doubt for the whole recommendation about whether you’re getting the best fit or just the company with the highest advertising budget.

In my actual experience, my AI chat included a local map view with some content cards for each electrical company in my area. So let’s take that card UI and use it as our starting point and ask a few ‘what ifs’:

• What if the ‘sponsored’ card showed the typical cost of the exact service I need, based on what other customers actually paid?
• What if there were coupons, discounts, or perks that directly benefit me if I contact a particular company?
• What if some of the “ads” weren’t just more electricians, but alternative routes like payment plans or DIY tutorials?

Now compare that conventional response to one that uses smartly integrated ads: the same electrician options, but with richer promotional features and a contextual panel that lives alongside the main thread instead of inside it.

Under the hood, this card layout is doing a few specific things. First, I kept the familiar basics — company name, review rating, a link to the website, a call-to-action (CTA) to get in touch, and a short description. Second, it uses information grouping: cost, reviews, and perks live together inside each card, so you don’t have to mentally stitch them across the interface.

Where it begins to differ is in the cost row: instead of a vague price range, the card shows the average amount customers actually reported paying for the specific service I’m asking about, plus a link to the reviews that generated that cost amount. That turns the “ad” element into a concrete expectation‑setting tool, not just a badge.

There’s also a clearly labeled coupon or perk add‑on that directly benefits the user as a form of promotional content, much like scouting through RetailMeNot or Groupon for a promo code to paste in at checkout, except the value is surfaced in context, before you ever hit a checkout page.

Last, it strengthens information scent by labeling actions in the user’s own language (“Reliable DIY tutorials,” “Electricians with payment plans,” etc.) instead of generic ad copy, which makes them feel like relevant next steps rather than distractions. Together, these tweaks improve the overall signal‑to‑noise ratio by making sure every sponsored element either clarifies trade‑offs or opens a concrete, optional path you might actually want to take.

Each predetermined CTA launches a focused sub‑thread where the AI can surface related content or even sponsored resources, without hijacking the main answer. This is where the “smartly integrated ads” pattern really shows up: the promotion lives in a side‑path you choose, rather than inside the core explanation you came for.

For the overall layout, all of this can be contained in a contextual, right‑aligned panel that fights banner blindness by keeping the promotional content inside a clearly related, text‑heavy space rather than in a noisy strip at the margins. It’s still clearly promotional, but it behaves like a helpful feature stitched to your thread instead of a banner shouting from the edge of the screen and avoids the frustration tax of random, off-topic ads.

Cards like these are probably the “maximal” version for a high‑stakes case, but even one or two of these ideas would give users a real sense of reciprocity and incentive to interact with the advertised picks. Companies willing to pay for ad space could compromise by allowing more transparency — e.g., showing the specific average amount customers paid, offering concrete perks and coupons, or sponsoring alternative routes like clearly labeled DIY content that gives the user more autonomy.

Compared to the tiny “Sponsored” tag next to BrightSpark Electrical in the earlier version, this card makes the advertising work much more in my favor. The promoted content is still an ad, but it’s wrapped in concrete information about cost, clear perks, and optional side paths that match my concerns, all reusing the same thoughtful pattern instead of casting doubt on the recommended services. That makes it more likely I’ll actually engage with it, instead of scrolling past it as just another banner or quietly wondering if the whole recommendation is spurious.

Mapping out the new frontier

Free‑tier AI platforms with ads don’t have to inherit the same frustration tax we’ve learned from YouTube commercials and sponsored search results. The electrician example is small, but it shows how promotional content can be reshaped into something that actually helps in a stressful, high‑stakes moment instead of quietly eroding trust.

Structurally, the move is simple: treat ads as enhancing features, not just paid recommendations. Smartly integrated cards, natural sub‑thread CTAs, and contextual panels give advertised content visibility while adhering to good principles like information grouping, strengthening information scent, and protecting the signal‑to‑noise ratio of the thread — which in turn counters banner blindness rather than fighting against it. At their best, these patterns also line up with the TRAP virtues I’ve argued for elsewhere: they make sponsorship transparent, offer real reciprocity in the form of perks and clearer information, and preserve user autonomy by keeping every sponsored path optional.⁶

As more frontier AI platforms roll out free‑tier advertising, the real question isn’t whether to show ads at all, but which patterns we normalize. Do we copy the flimsy “Sponsored” tag and hope users still engage with it, or invest in ad experiences that users would actually miss if we took them away? If you work on these products — as a designer, PM, or engineer — now is the moment to decide which pattern AI platforms ship.

References

[1] Simo, Fidji. “Our Approach to Advertising and Expanding Access to ChatGPT.” OpenAI. January 16, 2026. https://openai.com/index/our-approach-to-advertising-and-expanding-access.​

[2] Capoot, Ashley. “OpenAI to Begin Testing Ads on ChatGPT in the U.S.” CNBC, January 16, 2026. https://www.cnbc.com/2026/01/16/open-ai-chatgpt-ads-us.html.

[3] QQTube (Terri Pinyerd). “When Did YouTube Start Ads? | A History of Video Advertising.” QQTube Blog, October 6, 2025. Accessed February 9, 2026. https://www.qqtube.com/blog/when-did-youtube-start-ads.

[4] Holmes, Gareth. “A Brief History of Video Advertising.” New Digital Age, 2026. https://newdigitalage.co/technology/a-brief-history-of-video-advertising.

[5] GrowthSRC Team. “Banner Blindness Statistics & Studies You Need to Know in 2025.” GrowthSRC. Last modified 2026. https://growthsrc.com/banner-blindness-statistics-studies.

[6] Walsh, Tanner. “Eudaimonistic-Centered Design: The Virtues of UX.” UX Collective (Medium). February 16, 2022. https://medium.com/user-experience-design-1/eudaimonistic-centered-design-92b69654ee25.

Designing useful ads was originally published in UX Collective on Medium, where people are continuing the conversation by highlighting and responding to this story.