The AI answer is an ad

I used to trust Google the way I trust a dictionary. You look something up, you get the definition. No agenda. No commission. Just the thing.

Last week, Google published a blog post about their new ad formats for the AI era of Search. I read it through twice, and somewhere between “Conversational Discovery Ads” and “Highlighted Answers,” the dictionary started trying to sell me a fridge.

The story in one sentence

Google is building Gemini into the ads themselves - so the AI that writes your search answer can also write a personalized pitch for a sponsor, indistinguishable in format from an organic recommendation.

What they shipped

The announcement covers five new formats, and the interesting ones are the first two:

Conversational Discovery Ads are the strangest. When you type a long, conversational query - “I want my house to smell like a rainy forest, low maintenance options” - the Gemini model reads your query and writes a custom ad that answers it. The advertiser doesn’t write that copy. Gemini does. The advertiser just gets the click.

Highlighted Answers are more straightforward but arguably more damaging: when AI Mode produces a list of recommendations (say, “best language apps for a trip to Japan”), eligible high-quality ads can appear as items on that list, labeled “Sponsored.” You see a ranked list. Some entries are organic. Some are paid. You’re expected to notice the fine print.

Then come three more:

Format	What it does
AI-powered Shopping ads	Gemini writes a custom product explainer per search query
Business Agent for Leads	Instead of a form, you chat with a brand’s AI; Google hosts the agent
Direct Offers / Promotion Bundling	Gemini assembles custom deal bundles based on your search intent

The Direct Offers expansion is also real: Chewy, Gap, and L’Oréal have been in the pilot since January 2026. Travel partners like Booking and Expedia are next.

Why this hit 584 upvotes on Hacker News

Google has been threading this needle for years - making ads look like results - but text inside an AI-generated answer is categorically different from a link in a sidebar.

Adblockers work by targeting DOM elements: script tags, image URLs, known ad network endpoints. They can’t block a sentence that Gemini wrote inside the response body. That’s the shift. For twenty years, the technically literate had a clean exit route. This closes it.

The post hit the front page not just because of the content, but because of the implied announcement: AI Mode is the default Search experience now. The old plaintext web results are still there, but Google’s energy - and Google’s revenue model - is pouring into the conversational layer. And now that layer has ads in it.

What HN is arguing about

The most-upvoted comment comes from svieira, and it’s genuinely chilling:

“The particularly worrying thing here is that they’re now going to be gathering training data for a conversational model on how to influence people effectively even when they already know they are being influenced.”

This is the part that goes beyond the usual ad-fatigue complaint. RLHF already involves optimizing for human approval. Conversational Discovery Ads create a new feedback loop: Gemini writes ad copy, users click (or don’t), and Google now has training signal on what persuasive language works at every intent level, at scale, forever. That data doesn’t just improve ads. It improves the model’s ability to persuade you of anything.

karlkloss asked the quieter question: “Will the answers the AI gives also be influenced by Google’s customers?” Google’s blog post says no - the AI explainers are “independent” and run separately from the advertiser’s creative. That’s technically reassuring and psychologically useless. Every user already knows Google is an ad company. Seeing a “Sponsored” label next to a Gemini-generated recommendation doesn’t restore trust. It just makes the deal explicit.

neuropacabra announced they switched to DuckDuckGo after Google made AI Mode the default. Several others agreed, with varying levels of conviction. The FinnKuhn observation is worth sitting with: “Google is probably the player that could afford to wait the longest with this and increase their market share.” Meaning: they didn’t need to do this to survive. They chose to.

Read this if / skip if

Read the original if…	Skip if you…
You run Google Ads campaigns	Already switched to DuckDuckGo
You want to understand what “Business Agent for Leads” means for your pipeline	Haven’t touched Google Search in two years
You’re curious how Gemini writes ad copy differently than a human	Only care if adblockers still work (short answer: less so)

One thing that won’t change

Google’s stock reaction to criticism like this is to say the ads are “clearly labeled.” It’s true - they say “Sponsored.” It’s also true that banner blindness has been documented since 1998, and that the label on a Highlighted Answer means nothing once the AI has convinced you the recommendation is authoritative.

The implicit contract of search has always been: we’ll show you ads, you’ll learn to ignore them, nothing will be mixed. That contract just expired. What’s next is a model that knows exactly which sentence convinces you to order the espresso machine, and it will write that sentence, for a fee.

The dictionary is taking commissions now.

---

Discussion on Hacker News · Source: blog.google · Submitted by sofumel