ChatGPT ads or GEO: which delivers more?

Key findings

1. Cheap to start. Since 31 August 2026, ChatGPT ads can be booked self-serve in Austria, from EUR 15 per day per campaign.¹⁵
2. A smaller audience than you'd think. Only free and Go users see ads. Plus, Pro, Business, Enterprise, users under 18 and temporary chats stay ad-free.³ Without logging in, we saw not a single ad.⁴³
3. Ads don't buy a recommendation. In three US datasets, the advertiser also appeared in the answer above its own ad in only 2.8 to 5.4% of cases.³⁰³¹³²
4. Results so far are weak. In all four published cases where one company compared ChatGPT ads directly with Google Ads, Google delivered far more conversions per click.¹²
5. Several industries are blocked. Health, legal and financial services can't currently advertise in Europe, and individual property listings can't anywhere.³⁶
6. Organic takes longer but reaches everyone. The recommendation in the answer is also seen by paying users, who never get ads.

Paid and organic side by side

Our measurement: ads without login in Vienna

OpenAI says that users without an account can also see ads, limited to family-safe ones.³ Nobody had documented this for Austria. So on the evening of 13 October 2026 we asked ChatGPT nine typical recommendation questions without logging in, from Vienna, one per industry, each in a fresh chat. We checked every answer for ads, both on the page and in the data stream ChatGPT sends in the background. The questions were asked in German; the table shows them word for word.

For comparison: during an analysis for a client in September, a logged-in Austrian free account showed an ad below 5 of 7 B2B answers, from Austrian companies and in German. We don't publish those questions out of respect for the client.

What this means: in Austria, ads mainly reach people with a free account. People asking without an account saw only the organic answer in our test. But nine questions on one evening are a sample, not proof that there are never ads without login. Even logged in, ads don't appear all the time. In a German panel of 621 users, 41.2% saw at least one ad in the first month, on 17.7% of the days they used ChatGPT.¹⁰

One more observation: almost every answer ended with a clickable follow-up form, asking for district, budget or occasion, before ChatGPT got more specific. A business missing from the first answer can still appear in the second.

How ChatGPT ads work

OpenAI tested ads in the US from February 2026. On 18 August it announced 31 European markets, including Austria and Germany. Ads started rolling out from 24 August, and since 31 August companies can book their ads themselves in "Ads Manager".¹²

• Format: a labelled card below the finished answer, with name, logo, headline, text, image and link. Paid content does not appear in the answer text.⁴
• No keywords: OpenAI matches the ad to the conversation using the ad copy, the landing page and up to 2,000 free-text "context hints" per ad group.⁶ A dedicated crawler, OAI-AdsBot, checks every landing page.⁷
• Not personalised in Europe: in the EU, EEA and Switzerland, only the current chat, language and rough location count, not earlier chats or memories.³ OpenAI plans opt-in personalised ads. The data protection authorities of Hamburg and Berlin warned about this on 2 October.⁸
• No-go zones: ChatGPT shows no ads next to conversations about personal health, mental health or politics.³
• Price: an auction weighted by relevance. You pay per 1,000 impressions or per click, never per lead. OpenAI suggests starting bids of USD 3 to 5 per click and says it has no benchmarks yet.⁴ With a euro account the minimum is EUR 15 per day per campaign, averaged over a week. The currency is fixed when the account is created.⁵⁶
• Measurement: pixel, conversions API and UTM parameters. Reports show totals only, such as impressions, clicks and costs.⁴

OpenAI says its ad business reached a USD 1 billion annual run rate in under 200 days.⁹ At least 15 agencies in Austria, Germany and Switzerland now offer ChatGPT ads management, with published prices from about EUR 250 a month plus setup (our review of their pricing pages, October 2026).

What ChatGPT ads have delivered so far

Solid numbers are rare, and most come from companies that sell ads or tools. Still, the picture is fairly consistent.

Few clicks. In the US pilot, 0.91% of users clicked a ChatGPT ad according to Adthena, against 6.4% for Google search ads. Similarweb measures 0.73% for the US.¹¹ Early European campaigns saw higher click rates, on small budgets with little competition.¹²

Cheap clicks, but no market price yet. In ten European campaigns from published case reports, a click cost EUR 0.16 to 2.68.¹² That reflects a new auction with few bidders. Cheap clicks were not always good clicks.

Fewer conversions than Google. In all four published cases where the same company compared both channels, Google was clearly ahead:¹²

Two of the four cases come from an agency that sells ChatGPT ads itself. B2B reports are the weakest. SE Ranking tested for about two weeks at a 1.30% click rate and USD 3.16 per click, and got hardly any sign-ups.¹³ A German agency reports that in one test OpenAI recorded 36 conversions while the advertised company received only 2 real inquiries.¹⁴ The only repeated good results come from online shops, for example one shop that took 3.3 to 6.8 times its ad spend in revenue, depending on the attribution model, over 34 days on a USD 9,620 budget, measured by an agency that sells ChatGPT ads.¹²

Lots of testing, little budget. In a survey of 58 practitioners, 77.6% supported limited tests, but only 5.2% had moved existing budget into ChatGPT. 79.3% were worried about measurement.¹⁵

Who advertises. In US data, the market leader ran no ChatGPT ads in 12 of 14 industries. Advertisers are more often comparison sites, affiliates and challengers.¹⁶ Among about 15,000 companies with the ChatGPT ads pixel, hospitality and retail appear 2.3 to 2.4 times as often as expected, healthcare only 0.65 times.¹⁷

What the organic recommendation delivers

Organic visibility means ChatGPT names a business in the answer itself or cites its page as a source. All users see this, including those on paid plans who never get ads.

• Access is a must: sites that block the OAI-SearchBot crawler don't appear in ChatGPT's search answers. According to OpenAI, a robots.txt change takes effect after about 24 hours.⁷
• Official facts count for more: according to Ahrefs, 6.8% of ChatGPT's background searches target a single website. Among the roughly 160 sites searched this way most often, three quarters of these searches went to brand websites. Reddit fell from about one in five US answers in March to one in 43 in September.¹⁸
• It takes time: in a Semrush test with 81 new pages on a strong domain, ChatGPT cited 17% after 7 days and 42% after 30 days.¹⁹ Mentions in other sources take months.
• Few clicks: in Austria, 21% click through to a website after a chatbot answer, against 39% after a Google search.²⁰ ChatGPT brings websites 0.32% of their traffic on average.²¹
• Impact without a click: in a US panel, after a brand was named in ChatGPT, 6.39% of users visited its website within 7 days, against 4.33% without a mention. Over 97% of those visits carried no referrer, so analytics tools don't credit them to ChatGPT.²²
• Quality is disputed: the only peer-reviewed study, of 973 online shops, found a 13% higher purchase likelihood for visitors from organic search than for visitors from ChatGPT.²³ Adobe, by contrast, sees higher purchase rates for AI visitors than for other traffic in US retail.²⁴
• Volatile, but not random: the chance that ChatGPT gives the same list twice is under 1%. A brand's share across many queries stays fairly stable, though.²⁵ In our own Vienna measurement, answers overlapped 42.8% minutes later and 32.8% after 15 days.²⁶

Trust. All figures here come from the US. 63% of AI users trust AI recommendations, but most then check sources and reviews.²⁷ 41.6% of adults dislike ads in chatbots, and two thirds of them say ads make them doubt the answers.²⁸ No such survey exists for Austria.

Ads don't buy a recommendation

In January 2026 OpenAI promised that ads do not influence ChatGPT's answers and stay separate and labelled.²⁹ Nobody has tested this independently yet. But three US analyses are consistent with it:

³⁰³¹³² The reverse also holds: a recommendation doesn't protect the slot below it. On questions that named a brand and showed an ad, a direct competitor's ad appeared in at least 5.1% of cases, the brand's own ad in only 1.7%, and an unrelated company's ad in 93.2%.³²

What Google search teaches us. There is no data on this for ChatGPT yet, but there is for search engines. Google analysed 390 studies in which ads were paused: when the company ranked first organically, only half of the ad clicks were additional; at positions 2 to 4 it was 82%, and with no organic result 100%.³³ At eBay, after ads on its own brand name were switched off, 99.5% of the clicks came back through the normal search results.³⁴ At Bing, such ads added only 1 to 4% more clicks as long as no competitor was bidding. When competitors bid, a brand without its own ad lost 18 to 42% of its clicks.³⁵

Applied to ChatGPT, as an analogy and not a measurement: ads pay off where a business is missing organically. Where it is already recommended, you mostly pay for clicks you would get anyway. You only need to defend yourself once competitors advertise on the same questions.

The common ground is the website. OAI-SearchBot reads it for search, OAI-AdsBot for ads.⁷ A clear, factual page helps both channels. Blocking either crawler loses that channel.

Who is allowed to advertise in ChatGPT

The first hurdle is not the law but OpenAI's own ad policy. Version 1.6 of 10 September 2026 is the sixth change since March.³⁶

Some trade articles report case-by-case approvals for health and finance. Those refer to the US or to older versions. Check Ads Manager before every campaign.

Austrian law. A ChatGPT ad is ordinary advertising. Claims must be provable, so "Vienna's best estate agent" needs a verifiable source. An ad saying "ChatGPT recommends us" would be misleading and also breaks OpenAI's rule against imitating ChatGPT. Since 27 September 2026, generic environmental claims such as "eco-friendly" towards consumers are banned unless recognised excellent environmental performance is proven, and so is "climate neutral" when it rests only on offsetting.³⁷ Since 1 July 2026, every property ad must state heating demand, final energy demand and efficiency class, including the agent's.³⁸ The ad pixel may only load after active consent in the cookie banner.³⁹ Doctors, vets and lawyers may advertise factually, but without self-praise, comparisons with colleagues or paid referrals.⁴⁰

All ads will become public. The European Commission designated ChatGPT a very large online search engine on 31 August 2026.⁴¹ OpenAI must therefore keep a public ad repository with every ad, the advertiser, the payer, the main targeting parameters and the reach, probably from the turn of the year.⁴² Competitors will be able to see who advertises and how.

Organic has red lines too. An unlabelled paid spot in a "best of" list, posts where you pose as a customer, and bought reviews are always unfair under Austria's UWG, even if they only target AI recommendations.³⁷

Which strategy for which industry

The pattern: ads fit where customers use ChatGPT for free and buy or book online. For most local professions, ads are either prohibited, hard to measure or too broadly targeted. For them, the recommendation in the answer remains the only way into ChatGPT.

What businesses should do now

• First check whether ChatGPT already names you, with repeated questions, not a single test.
• Keep your website open to both crawlers: OAI-SearchBot always, OAI-AdsBot as soon as you plan ads.
• Only test ads if three things hold: your industry is allowed, your customers use ChatGPT for free, and inquiries can be measured in your own system. For a fair test, plan on 6 to 8 weeks and about EUR 450 ad budget a month.
• Measure success by real inquiries and sales, not by OpenAI's conversion count.
• Don't believe any guarantee, neither for "number 1 in ChatGPT" nor for ad results. Answers change with almost every query.
• Read rules with a date: what is prohibited today may be allowed next month.

What nobody knows for sure

• Whether individual cities such as Vienna can be chosen as a target area in Austria. The sources contradict each other.
• Whether vets and tax advisors may advertise.
• Whether users without login in Austria ever see ads.
• How often ads appear in Austria at all. All frequency studies come from the US.
• Whether OpenAI's promise that ads don't influence answers holds. Nobody has tested it independently.
• What effect GPT-6 has, the free version's model since 8 October 2026. There is no independent data on it yet.

Quick answers

Method

Type

Analysis of published sources plus our own sample in ChatGPT

As of

13 October 2026

Sources

More than 200 reviewed: OpenAI documents (via archive.org, because openai.com blocks automated access), laws from RIS and EUR-Lex, peer-reviewed studies, surveys, data from tool vendors and agencies. The most important ones are linked below.

Checking

Before publishing, we checked every key figure a second time against the original source and corrected inaccurate details.

Own measurement

13 October 2026, about 23:15 to 23:45, Vienna, ChatGPT without login (free model), 9 questions, each in a fresh chat. Checked: page and answer data stream for ads.

Labelling

Figures from companies that sell tools or ads are marked as vendor data in the source list. US-only data is named as such.

Limits

There is hardly any data for Austria; most figures come from the US. Nine questions on one evening are a sample.

Sources

1. OpenAI, ChatGPT ads expands across Europe (18 August 2026, update 31 August 2026), archived copy.
2. ORF, report on the start of ChatGPT ads in Europe (August 2026).
3. OpenAI Help Center, Ads in ChatGPT, archived copy of 19 September 2026.
4. OpenAI Help Center, Ads in ChatGPT: the basics, and FAQ, archived copies October 2026.
5. OpenAI Help Center, Create campaigns for ChatGPT ads (minimum budget), archived copy.
6. OpenAI Ads developer documentation (context hints, account currency), accessed 13 October 2026.
7. OpenAI, Overview of OpenAI Crawlers (OAI-SearchBot, OAI-AdsBot), accessed 13 October 2026.
8. Notebookcheck, data protection authorities warn about profiles from chats (October 2026, in German).
9. OpenAI, Expanding access to AI with ChatGPT ads (OpenAI's own run-rate figure), archived copy.
10. marktforschung.de, Murmuras panel Germany, 621 users, 27 August to 27 September 2026 (in German).
11. The Keyword, Adthena data on the US pilot; plus PPC Land on Similarweb and Graphite.
12. PPC One, collection of 49 published case reports on ChatGPT ads (October 2026). Vendor data.
13. SE Ranking, own test with ChatGPT ads (2026, in German). Vendor data.
14. OMR, ChatGPT Ads: first experiences (2026, in German).
15. OneMetrik, survey of 58 practitioners (September 2026).
16. Otterly, ChatGPT ads study (2026), US data. Vendor data.
17. Bloomberry, Who's buying ChatGPT ads, about 15,000 companies with the ChatGPT ads pixel (2026). Vendor data.
18. Ahrefs, How ChatGPT Does Query Fan-Out (12 October 2026), 66.6 million free logged-out answers. Vendor data.
19. Semrush, How fast do AI search platforms cite new content (2026), 81 new pages. Vendor data.
20. Leadersnet, WPP Media study Austria, 1,032 online interviews, ages 18 to 65, July 2026, self-reported (7 October 2026, in German); also reported by Der Standard.
21. SE Ranking, ChatGPT referral traffic, May 2026, about 100,000 websites. Vendor data.
22. Profound, The AI mention effect (2026), US panel. Vendor data.
23. Marketing Science 45(4), ChatGPT Referrals to E-Commerce Websites: How Do LLMs Compare Against Traditional Channels? (peer-reviewed, open access, 973 online shops, August 2024 to July 2025).
24. Digital Commerce 360, Adobe data on AI referrals in US retail (19 August 2026).
25. SparkToro, New research: AIs are highly inconsistent when recommending brands or products (28 January 2026).
26. KIGEO, Wie verlässlich sind KI-Empfehlungen? (How reliable are AI recommendations?, 6 August 2026, in German), own measurement.
27. BrightLocal, survey on trust in AI recommendations, 455 AI users among 1,002 US adults (2026).
28. Semrush / Exploding Topics, survey of 2,338 US adults (2026). Vendor data.
29. OpenAI, Our approach to advertising and expanding access (16 January 2026).
30. SE Ranking, ChatGPT ads study, 50,006 prompts, US (July 2026). Vendor data.
31. Seer Interactive, ChatGPT ads insights, US (July 2026). Vendor data.
32. LLM Pulse, ChatGPT ads study, 34,040 ads, US (July 2026). Vendor data.
33. Google Research, Impact of Ranking of Organic Search Results on the Incrementality of Search Ads (390 studies).
34. Blake, Nosko, Tadelis, Consumer Heterogeneity and Paid Search Effectiveness: A Large-Scale Field Experiment (Econometrica 2015), eBay experiment.
35. Simonov, Nosko, Rao, Competition and Crowd-Out for Brand Keywords in Sponsored Search (Marketing Science 2018).
36. OpenAI, Ad Policies, version 1.6 of 10 September 2026, archived copy.
37. RIS, Austrian Unfair Competition Act (UWG), current version.
38. RIS, Energy Performance Certificate Act (EAVG), § 3.
39. RIS, Telecommunications Act 2021, § 165.
40. RIS, Physicians Act § 53; plus Veterinarians Act § 31 (jusline.at) and ÖRAK, Guidelines for the Practice of Law 2015, §§ 47, 48.
41. European Commission, designation of ChatGPT as a very large online search engine, IP/26/1772 (31 August 2026).
42. EUR-Lex, Digital Services Act (DSA), Regulation (EU) 2022/2065, Art. 39.
43. KIGEO, own measurement: 9 questions, ChatGPT without login, Vienna, 13 October 2026 (this page).