Bulk campaigns, virtual try-on, and a tracking problem for ChatGPT ads

ChatGPT is starting to look less like a chatbot and more like a storefront.

Building the store: OpenAI just rolled out a batch of ChatGPT Ads updates, mostly for e-commerce:

• Bulk product-feed campaigns from a spreadsheet upload. Ads Manager builds the ad templates for you.
• Product review status that shows why OpenAI rejected individual products.
• Conversions beyond your campaign goal, pulled through the Insights API.
• New markets across Southeast Asia and Taiwan.

OpenAI says product-feed ads perform best on the platform so far, so catalog-heavy brands have the most reason to test.

Now stocking the shelves: ChatGPT shopping results also picked up virtual try-on and Favorites globally. Shoppers upload a selfie, hit Try On, and save what they like.

That moves ChatGPT onto Pinterest and Google turf for style discovery. OpenAI already shelved Instant Checkout after it underperformed, though.

The receipt is the problem: Agencies say measurement gaps keep budgets at test level.

One saw zero platform conversions two weeks into a lead-gen run, and about 100 reported clicks against 20 it tracked.

Others see conversions land 24 to 36 hours late, with no demographic or prompt data. Many legal teams also refuse to sign OpenAI's tracking terms.

So set up your own measurement before you test, and judge results by your own data.

Until OpenAI's numbers catch up, keep ChatGPT spend in the test column.

82% of branded searches now get an AI Overview, and publishers get pocket change

Nobody asked for a narrator, yet Google cast itself as Morgan Freeman for your brand anyway.

Say my name, say my name: AI Overviews jumped from 26% to 82% of DemandSphere's tracked branded queries in September, peaking at 90% on Sept. 27.

Google hasn't announced a thing.

Before you panic, most branded AI Overviews sit mid-page, below the official site. Only Google's own name and news publishers seem exempt. Convenient.

Still, the summary appears whether it cites you or not. So check:

• Which branded queries trigger one.
• What it says about you.
• Which sources it cites.

Speaking of citations, Google's AI contribution pilot pays some publishers under 0.1% of their ad revenue. Payouts appear in Search Console with zero explanation.

So don't build your content plan around AI licensing checks.

The ad tech bill might come due, though: A federal judge let publishers' core antitrust claims proceed against Google.

Experts for Gannett, Daily Mail and a publisher class put damages above $3B.

If you buy programmatic, the fees and auction rules in your stack are heading to trial.

Google may end up paying publishers real money yet, courtesy of a judge.

Question titles are costing you clicks

We almost wrote this headline as a question.

Then we read the latest breakdown at Science Says and thought better of it.

The gist: A new Stanford study shows question-style titles cost you clicks, comments, and credibility. Here’s why.

1) The numbers don’t ask: Researchers Fang and Wheeler ran 3 experiments and analyzed over 22,000 news articles, 3,000,000 academic papers, and 53,000 Reddit posts.

Compared to statements, question titles earned 71.6% fewer comments, 53.7% fewer upvotes, 12.6% fewer clicks, and 4.9% fewer citations.

2) Rhetorical, direct, open-ended: all equally guilty: The type of question made no difference. Every flavor dragged engagement down.

So no, the "clever" rhetorical question doesn't get a pass. We checked our own archive. We might have some rewriting to do.

3) Certainty gets clicks, doubt gets scrolled past: Readers hunt for clear, certain information when they decide what to open.

A question signals uncertainty, so readers judge the content as less informative and less valuable before they even click.

4) Swap the format, keep the topic: The study points to Zapier's "App comparisons" collection, which leans on question titles.

The fix: adopt a repeatable statement format like "X compared to Y" and add assertive language.

Also, add more spine. Turn "Do you need to optimize for AEO?" into "When you need to optimize for AEO."

Before you bulk-edit old titles, confirm the change doesn't hurt your SEO.

5) The fine print still applies: The study covered Western media, and the effect may shift when people read for leisure.

It could hit even harder in uncertainty-avoidant cultures like Central Europe and Latin America.

Since this is new research, test it on a small batch of titles before you rewrite your whole content library.

You can read the full breakdown, including the real-life Zapier example, here.

Claude gets people talking… a lot

Some chats end with "thanks." Others turn into a three-hour coffee date.

A new study by Comscore counted how many prompts people send per conversation, and one AI assistant gets a lot more back-and-forth than the others.

Here's the average number of prompts per conversation on the big four assistants:

• Claude: 9.1 prompts, about 69% more than ChatGPT.
• Microsoft Copilot: 6.9, just ahead of Gemini at 6.2.
• ChatGPT: 5.4, last place despite having by far the most users.

Note that longer chats aren't automatically better. They can mean deeper work, like coding, writing, and analysis. Or they can mean it takes more rounds to get the answer right.

ChatGPT's shorter sessions probably come from its huge, casual user base treating it like a search engine: quick question, quick answer, done.

What we think: It comes down to what people use each tool for. Claude is a favorite for coding, writing, and long projects that need lots of rounds. ChatGPT is everyone's quick-lookup tool.

Actionable tip: When you plan for AI search, don't stop at the first question.

Also create content for the follow-ups people ask next, like comparisons, pricing, and how-to steps. That's where users who send lots of prompts make their decisions.