ChatGPT AdsChatGPT AdvertisingAI AdvertisingOriginal ResearchAI Search Visibility

ChatGPT Ads Study (November 2026): 2 in 3 Answers Had an Ad, and 4 in 5 Advertisers Weren't in the Answer

We asked logged-out ChatGPT 100 buying questions, then repeated 40 of them twice. 66 of the first 100 answers carried an ad. Across 136 ads, ChatGPT named the advertiser in its own answer 29 times. Per-category results, how often ads change, and the full data.

Nov 14, 2026
RankScope Team
Share:
ChatGPT ads study, November 2026: 66 of 100 answers had an ad, the advertiser was named in the answer for 29 of 136 ads, and the advertiser changed between runs for 21 of 30 questions

TL;DR

  • On 14 November 2026 we asked logged-out ChatGPT, in the US, 100 'best X for Y' buying questions across 10 categories. 66 of the 100 answers carried an ad under them.
  • Across all 136 ads we recorded (192 answers, including repeats), ChatGPT named the advertiser somewhere in its own answer 29 times. 107 times the ad offered something the answer had not mentioned.
  • Software is the exception. In B2B software the advertiser was in the answer for 10 of 19 ads (Bill.com, Bitwarden, Pipedrive, QuickBooks). Outside software and personal finance it was 13 of 101.
  • About a quarter of ads (37 of 136) were bought by middlemen: publishers, comparison sites, directories, lead-gen and booking sites. Forbes and Expedia ran 9 ads each.
  • Ads move. We asked 40 of the questions three times that afternoon: 21 had an ad every time, 13 only sometimes, 6 never. Where a question showed an ad at least twice, the advertiser changed in 21 of 30.
  • Finance and health questions were not ad-free: 9 of 10 finance and 6 of 10 health questions carried an ad in the first round. Consumer electronics had none in the first round and 2 in 8 repeats.
  • This is one afternoon from one US location, logged out. It describes what we saw, not how often ChatGPT shows ads overall. All 192 answers are in a free CSV.

When ChatGPT shows an ad under a buying question, is the advertiser a brand the answer itself recommends? Usually not. We asked logged-out ChatGPT 100 buying questions on 14 November 2026, then repeated 40 of them twice more. Out of 136 ads, ChatGPT named the advertiser in its own answer 29 times.

This page has the findings, the method, the hypotheses we wrote down before collecting anything (including the ones that failed), and the full data.

New to ChatGPT ads? Start with how ChatGPT ads work and what they cost.


Key Findings

  • 2 in 3 answers carried an ad. 66 of the first 100 answers had an ad below them.
  • 4 in 5 advertisers weren't in the answer. For 107 of 136 ads, the answer above never mentioned the advertiser.
  • Software is the exception. In B2B software, the advertiser was in the answer for 10 of 19 ads. Category leaders are buying the slot under answers that already recommend them.
  • A quarter of ads come from middlemen. 37 of 136 ads came from publishers, comparison sites, directories, lead-gen and booking sites.
  • Finance and health were not ad-free. 9 of 10 finance questions and 6 of 10 health questions carried an ad in the first round.
  • The same question rarely shows the same ad. Where a question showed an ad at least twice, the advertiser changed in 21 of 30.

Which Categories Carry the Most Ads

Bar chart: answers with an ad out of 10 per category, first round. Education 10, marketing tools 9, travel 9, personal finance 9, local home services 8, pets and food 7, health and wellness 6, B2B software 5, ecommerce and DTC 3, consumer electronics 0

CategoryFirst round (of 10)Repeats with an ad
Education108 of 8
Marketing tools914 of 14
Travel95 of 8
Personal finance97 of 8
Local home services88 of 8
Pets and food74 of 8
Health and wellness65 of 8
B2B software514 of 14
Ecommerce and DTC33 of 8
Consumer electronics02 of 8

Repeats: answers from rounds 2 and 3. Most categories have 4 repeated questions × 2 rounds. B2B software and marketing tools have more because round 2 covered all their questions before we cut the repeats to 40 (see Method).

Consumer electronics stands out. Not one of 10 first-round answers carried an ad, and only 2 of 8 repeats did: Back Market for "best budget smartphone", then Xfinity Mobile for the same question.

B2B software went from 5 of 10 in the first round to 14 of 14 in the repeats. We asked the same questions in the same way, about an hour later. That is the clearest sign in the data that ad delivery shifts within a single afternoon.


Is the Advertiser in the Answer?

Mostly no. For 107 of 136 ads, ChatGPT's answer recommended other brands and never mentioned the advertiser. Clover ran under an answer that recommended Square, Zoho and Lightspeed for inventory. DataCamp ran under a list of freeCodeCamp, Codecademy and Harvard CS50. Plenty of ads were tangential: a sheet-music library under a piano-app question, an anti-barking pendant under a dog-training-app question.

WhereAdvertiser named in the answer
B2B software10 of 19 ads
Personal finance6 of 16 ads
All other categories13 of 101 ads
Total29 of 136

Software leaders advertise against answers that already name them. Bill.com appeared under accounting and invoicing answers that mentioned it. Bitwarden, Pipedrive and QuickBooks ran under answers that listed them as options. The advertiser was the answer's first pick only 3 times (SoFi, Hostinger and Rocket Money). Mostly, the ad promoted a brand the answer had listed further down.

Does advertising change the answer? We didn't test that, and our data can't show it. OpenAI's stated policy is that "Ads do not influence the answers ChatGPT gives you".


Who Buys ChatGPT Ads

AdvertiserAdsType
Forbes9Publisher (ranked lists)
Expedia9Booking site
Bill.com4Product
Angi4Lead-gen marketplace
Acely4Product (SAT prep)
SITE123, Pimsleur, Rocket Money, QuickBooks3 eachProduct

37 of 136 ads (27%) came from middlemen rather than the product itself: publishers, comparison and review sites, directories, lead-gen services and booking sites. Forbes ran its "best of" lists under meal-kit, moving, credit-card, pet-insurance and pet-pharmacy questions. Angi and Thumbtack ran under local-service questions. Expedia ran under flight, hotel, vacation-rental and car-rental questions.

In the first round alone we counted 53 different advertisers across 66 ads.


Finance and Health Were Not Ad-Free

We expected fewer ads under personal-finance and health questions. We were wrong. In the first round, 9 of 10 finance questions and 6 of 10 health questions carried an ad. Advertisers included SoFi, Credible, Webull, Zocdoc and a protein-powder brand.

The framing was different, though. Above most ads, ChatGPT writes a short line tailored to the question, such as "For freelancer invoicing, here's one app option." For 9 of 136 ads it used a generic line instead, "This might also interest you." 8 of those 9 were under finance or health questions.

Some advertisers never get a lead-in. Every Forbes and every Expedia ad (9 each) appeared with no line above it, as did QuickBooks three times. This looks like a per-advertiser format rather than a per-question one. Ads are also localized: Expedia headlines named our own city, not one from the question.


Same Question, Different Ad

Stacked bar: of 40 questions asked three times, 21 had an ad every time, 13 some of the time, 6 never. When a question showed an ad in two or more runs, the advertiser changed in 21 of 30

We asked 40 of the questions (4 per category) three times, 35 to 96 minutes apart. Ads moved a lot:

  • "best social media scheduling tool": Wix, then SocialBee, then Monday.com.
  • "how do I find a good plumber in Austin": Plumbers Today, then PlumberProNow, then Plumbers Today again.
  • "best CRM for a small business": no ad, then Monday.com, then Pipedrive.
  • "best credit card for travel rewards": Forbes, then no ad, then BestMoney.

One check isn't a measurement, for an advertiser or for anyone tracking competitors. A brand that appears under an answer once may not appear the next time, and the same is true of the answer itself (why ChatGPT answers vary).

Screenshots of 8 of these ads, with the answer above each and a table of all 108 distinct ads, are on our ChatGPT ads examples page.


What This Means If You're Deciding Whether to Advertise

  1. The ad doesn't put you in the answer. For 4 in 5 ads, the advertiser sat under an answer that recommended someone else. Being in the answer is earned through the sources ChatGPT cites (how to get cited by ChatGPT).
  2. Check the answer before you buy the slot under it. If ChatGPT already recommends you, an ad may buy a click you'd get anyway. Software leaders are doing exactly that. If it doesn't, the ad is competing against the brands it does name.
  3. You may be bidding against middlemen. In local services, travel and finance, Angi, Expedia, Forbes and comparison sites are in the same auctions.
  4. Measure over time, not once. Ads and answers both change within the hour.

If you decide to test it, our setup guide covers context hints and the traps to avoid. Our free AI visibility checker shows whether ChatGPT names you for a question, with no account needed. RankScope tracks it daily across ChatGPT, Google AI Overviews, Google AI Mode and Perplexity.


Method

Questions. 100 buying questions, 10 in each of 10 categories, phrased "best X for Y" or "how do I choose X", none naming a brand. They were written and frozen before the first run.

Collection. chatgpt.com in a desktop Chrome browser, logged out, one US location, a new chat for every question. Four tabs ran in parallel. We read each page at least 30 seconds after sending the question, because ads sometimes load after the answer. An answer counted as having an ad when a card labelled "Ad" with a "Learn more" link appeared below it. For each ad we recorded the advertiser, headline, ChatGPT's lead-in line and landing domain, plus the numbered items in the answer.

Advertiser named in the answer. An ad counts when the advertiser's brand appears anywhere in ChatGPT's answer text. This was matched automatically, ignoring suffixes like "Inc." and "LLC", and then checked by hand. Five matches were corrected, such as "Intuit QuickBooks" against "QuickBooks Payroll".

Advertiser type. Classified by hand from the landing domain as either the product itself or an intermediary (publisher, comparison or review site, directory, lead-gen service, booking site or marketplace).

Timing. Round 1 ran 13:45 to 14:18 Pacific on 14 November 2026, round 2 from 14:20 to 15:24, and round 3 from 15:56 to 16:05.

What we predicted, before collecting any data:

HypothesisResult
The advertiser is named in the answer for 25% of ads or fewerHeld. 11 of 66 in round 1 (17%), 29 of 136 overall (21%)
At least 30% of ads come from publishers, directories or lead-genNot met. 17 of 66 in round 1 (26%), 37 of 136 overall (27%)
Finance and health questions show fewer adsWrong. 9 of 10 finance and 6 of 10 health questions had an ad in round 1
Ad presence and advertiser vary between runs of the same questionHeld. 13 of 40 questions changed between ad and no ad. The advertiser changed in 21 of 30

Changes to the plan, and why. We planned three full runs on three different days. Before the first run we moved all three to one afternoon. Logged-out ChatGPT then stopped accepting messages partway through ("You have reached the anonymous message limit"). So before collecting more, we cut rounds 2 and 3 to a fixed subset: the first 4 questions in each category. Round 2 had already covered all 20 B2B-software and marketing-tool questions by then. Those 12 extra answers are in the data and marked as outside the repeat set. We waited out every limit and did not work around it.

Limits. This is one afternoon, one location, logged out. Ads for logged-in Free and Go users can use personalization signals we didn't have. We recorded the first ad under each answer only. These are counts of what we saw, not rates for ChatGPT as a whole. Use them that way.


The Data

Download the CSV, with one row per answer (192 rows). It is free to reuse under CC BY 4.0 with a link to this page.

ColumnMeaning
run, question_id, category, questionWhich question, in which round
captured_utcWhen we read the page (blank for the first 4 rows, which we logged by local minute only)
ad_shown1 if an ad appeared below the answer
advertiser, ad_headline, chatgpt_leadin, landing_domainThe ad as shown. Three headlines named the city or state we ran the test from; we replaced those words with [our city] or [our state]
advertiser_named_in_answer1 if ChatGPT's answer mentioned the advertiser
advertiser_typebrand or intermediary
in_repeat_subset1 for the 40 questions asked in all three rounds
answer_top_picksNumbered items in the answer, in order. These can include follow-up questions ChatGPT asked

We plan to re-run all 100 questions in late November to measure week-over-week change, and will update this page with the date.


Frequently Asked Questions

How often does ChatGPT show ads?

In our test, 66 of 100 logged-out answers to buying questions carried an ad, on 14 November 2026 from one US location. That is what one afternoon looked like, not a rate for ChatGPT overall. It varied by category, from 10 of 10 education questions to 0 of 10 consumer electronics questions.

Are ChatGPT advertisers recommended in the answer?

Usually not. Across 136 ads we recorded, ChatGPT named the advertiser in its own answer 29 times. The exception was B2B software, where the advertiser was in the answer for 10 of 19 ads. OpenAI says ads do not influence the answer.

Do ChatGPT ads change each time?

Often. We asked 40 questions three times on the same afternoon. 21 had an ad every time, 13 only some of the time and 6 never. Where a question showed an ad at least twice, the advertiser was different between runs in 21 of 30 cases.

Who advertises on ChatGPT?

In our data, mostly product brands, plus a large group of middlemen. 37 of 136 ads came from publishers, comparison sites, directories, lead-gen services or booking sites. Forbes and Expedia ran the most ads, 9 each, followed by Bill.com, Angi and Acely with 4 each.

Can I use this data?

Yes. The CSV of all 192 answers is free to reuse under CC BY 4.0, with a link back to this page.

Related Articles