• by Laurie Sullivan , Yesterday

There is a disconnect in OpenAI ChatGPT's query answers when ads serve up in the engine.

About 14.35% of ChatGPT ads had no topical connection to the prompt, according to data released this week.

In niche categories like Relationships and in News & Politics, for example, the disconnect jumped above 50%, still running off topic. About one in seven ads has no topical connection to the prompt it appeared in. 

Advertisers appeared as a cited source in ChatGPT's answer to queries in just 3.63% of ad placements, compared with 11.53% in Google AI Mode.

Many advertisers tried out the ad system. SE Ranking, which conducted the study, found 1,159 unique advertisers -- including BestMoney, a fintech platform operated by Natural Intelligence, which invested heavily in OpenAI's ad format.

Alone, BestMoney accounted for 13.56% of all ChatGPT ad units.

While ChatGPT had no problem finding AI conversations to place ads, deliver traffic, and generate clicks at a reasonable rate, the real challenge came after the click, according to SE Ranking, a search agency that tested OpenAI's ad platform.

SE Ranking analyzed 50,006 commercial prompts across 20 niche markets in the U.S., using the same set of prompts from its previous AI Mode ads research.

For every prompt recorded, the system cited sources and a sponsored ad whenever one became available.

During the test, the agency also ran its own ChatGPT Ads to see what the platform looks like from the advertiser side. During the two weeks the ads ran, the campaigns generated more than 97,000 impressions and 1,263 clicks.

The biggest challenge from those impressions and clicks, the SE Ranking received very few sign-ups.

Described in the findings, the agency detailed how to set up the ads, along with results received from them and key limitations it encountered. It also noted that ads served in one in four commercial queries.

For two weeks, SE Ranking ran three campaigns, eight ad groups, and 48 ads across the U.S., Canada, Australia, and New Zealand.

The campaigns focused on people looking for AI visibility and AI Overview tracking tools; agencies interested in rank tracking, reporting, and generative engine optimization or answer engine optimization services; and technical users searching for SEO APIs and MCP integrations.

The biggest problem with results is that the campaigns generated very few sign-ups, so the results were not strong enough for the agency to continue. The blog cited "audience eligibility" as a possible reason. 

ChatGPT ads appear only to users on the Free and Go plans, while Plus, Pro, Business, and Enterprise users remain ad-free.

That model creates a challenge for agencies like SE Ranking because its "ideal customers" are agency owners, SEO leads and in-house specialists. These companies are heavy AI users and are more likely to pay for an ad-free plan.

"Attention on ChatGPT is cheap for our niche compared with the other paid traffic sources we use," wrote Vadym Hryshchenko, paid search specialist at SE Ranking, in the blog. "The problem is who's behind those clicks: ads only reach Free and Go users, and the more someone looks like our ICP, the more likely they’re on a paid, ad-free plan."

ChatGPT also distributes delivery of the ads "unevenly," meaning that the two largest ads received around half of all impressions, while some others received very little traffic.

One or two creatives inside an ad group often received most of the impressions, while the rest barely ran.

This means that conversation volume and the way ChatGPT interprets context hints play a major role in deciding where delivery goes.

The test also showed important limitations, noted during the platform's early days when manual cost per click (CPC) bidding first became available. There was no in-platform conversion data available for optimization.

Several features have been added since then, according to the blog.