ChatGPT Ads CTR by Market: How to Read the Benchmark

United States 0.73%, Brazil 1.91%, Mexico 1.55%, Japan 1.53%. Those are the all-time ChatGPT Ads CTRs by country that Graphite and Similarweb published on September 28, 2026 in The State of ChatGPT Ads, a 32-page report built on Similarweb panel data from the nine markets where ChatGPT serves ads, covering March 30 to September 16. Search Engine Journal has already covered it. If you buy ChatGPT Ads in the US, expect that table to show up in a budget conversation as proof that US users click less, or as the reason to shift a test to Brazil or Mexico. The report never says when each market starts counting, and without that, the table supports neither reading.

A cumulative table with no start date per market

The report doesn't state when each of the nine markets enters its data. What it does print is that ChatGPT officially launched ads in March 2026 and that its data series, US included, start on March 30. Over that period, according to the same report, worldwide weekly CTR rose 112%, though it doesn't explain how that figure is calculated: not which weeks it compares, nor whether it's a trend line or first week against last.

Brazil's timeline comes from outside the report. According to press coverage of an OpenAI briefing (Meio & Mensagem, propmark and Exame, August 4), ChatGPT started serving ads to users in Brazil on August 4, in an initial phase with agencies. That's press coverage of a briefing, not a company document, so treat it as unverified. On August 5, an OpenAI email to advertisers said: "In the coming week, we will launch ChatGPT Ads in Brazil and Mexico". Neither date tells you when Similarweb's panel started counting Brazil or Mexico, and those sources give no start date for Mexico. The difference between opening an account, serving ads and being able to buy in those two markets is laid out in our piece on ChatGPT Ads arriving in Brazil and Mexico.

Put together, the table lines up markets with very different time in market, over a period in which worldwide CTR rose 112% by the report's own count, and the report doesn't separate one from the other. A cumulative CTR depends on which weeks it adds up, so the US 0.73% and Brazil's 1.91% may be summarizing different moments of the channel. That means the table doesn't show that US users click less, and it doesn't work as a benchmark to project a campaign in any of the nine markets. It doesn't place the US above or below Brazil in comparable terms either: with what's published, that comparison stays open.

Who wrote the report, and what it says about its own data

Both companies behind the report have a stake in the topic, and that's worth knowing before you use any of its numbers. Graphite describes itself on page 4 as a growth agency that runs SEO, AEO and paid search, and the report closes with a button to its ChatGPT Ads playbook. Similarweb sells data, including the panel the CTR figures come from. That doesn't invalidate the numbers, but it helps explain why nearly every recommendation in the report pushes toward testing, benchmarking and iterating.

The report states its own limits on page 6: Similarweb discloses percentages and shares only, with no panel volumes; cross-channel CTR comparisons are limited to paid search; and Section 1 is "early and directional". It also has no spend, CPC, CPM, conversions, no definition of CTR as the panel measures it, and no weekly values behind its charts. Everything it says about advertisers, categories, creatives and landing pages comes from US samples, which makes that part of the report more relevant to a US buyer than to anyone else.

Turns, copy and landing pages, with the sample in view

The most usable data point in the report is which conversation turn the ad shows up in, because it comes from the panel with all countries combined. 51% of impressions land in the first 3 turns (all countries, March 30 to August 23), and per the report, ads in the first 4 turns get 48% higher CTR (worldwide, March 30 to August 30), without saying which group that's measured against or how it's calculated. The report defines turn 1 as the opening exchange, the first user message and its reply. It doesn't say whether advertisers can see or choose the turn their ad appears in, so that's worth asking the platform before you build a hypothesis around the number.

Copy and landing page data come from US samples. Across 10 US advertisers running on both ChatGPT and Google Search, only 1.9% of ChatGPT headlines exactly match their paid search headline, which is where the report's 98% net new copy figure comes from. Across 400+ US advertisers (July 19 to August 15), 60% use 5 or fewer landing pages. Ten advertisers aren't the market, but if you're already running Google Search in the US, both figures are a reasonable starting point for a test: copy written for ChatGPT rather than reused from search, and a short list of landing pages, is what most of the sample does.

What to log from week one

We recommend using the report to frame questions and setting targets on your own data, starting with a weekly log from the first day of the campaign. Record the campaign start date and, week by week, impressions, clicks and your own CTR, split by placement or by turn if the platform reports it, which is worth confirming before launch. Next to that go cost and conversions measured on your site or in your CRM.

Cost and conversions are exactly what the report can't give you: it has no spend, CPC, CPM or any post-click metric, so a high CTR, yours or the table's, says nothing about whether the click was worth it. It doesn't publish volumes either, so it can't tell you how many impressions you need before a week of your own CTR stops being noise; that threshold comes from your own account volume. For what to set up on day one, see our piece on conversion optimization, which OpenAI calls its recommended model, and for reading results outside the platform, how to measure ChatGPT ads.

If someone brings the 0.73% to argue the US is underperforming, or the 1.91% as a target for a Brazil or Mexico test, the useful question is when each market starts counting in that number and which of your own weeks it's being compared against. Until Graphite and Similarweb publish a start date per market, we recommend not using the table as a campaign target and comparing each week of your own data against the earlier weeks of the same account.

This content was developed with AI assistance and reviewed by the Zenda team. Any bad ideas are entirely our own.