Are bots clicking my ChatGPT ads, and will OpenAI credit me for invalid clicks?

TofuAnswer by the Tofu team · checked

A jointed wooden mannequin hand with one finger bent down as if pressing a button, on a dark navy background.

Some may be, and OpenAI's developer docs describe no credit for them. They say Clicks and Conversions campaigns bill "per valid click," but don't say how OpenAI spots invalid clicks, and offer no invalid-click report or claim process. In a study by TrafficGuard, which sells click-fraud protection, 0.1% to 34% of each advertiser's ChatGPT ad clicks were invalid, about 4% overall. To check yours, tag your clicks, compare OpenAI's click counts with your server logs and look for cloud-hosting IPs. Send any evidence to OpenAI's ads support quickly.

Ask Tofu next

Pick a question and Tofu Ads Agent answers it in chat.

As of November 2026.

Before you start

  • Your server or CDN logs for your landing pages, with the time, IP address, user agent and full URL of each request. GA4 alone won't do; Step 3 explains why.
  • Access to ChatGPT Ads Manager, plus an Advertiser API key if you want to pull clicks by ad and day through the API. You create the key on the Settings page in Ads Manager.
  • Your invoices or charge IDs for the period you want to question.

Step 1: Know what OpenAI bills, and what it doesn't promise

OpenAI's developer docs set billing by campaign objective:

Campaign objectiveWhat you pay for
ReachEach 1,000 impressions
ClicksEach valid click
Conversions"Per valid click, not per conversion"

That's all they say. The docs don't define a valid click, explain how OpenAI filters invalid ones, or offer a report of filtered clicks or credits. PPC Land, reporting on the TrafficGuard study in Step 4, notes that Google documents an Invalid Activity Credit report and that the study's materials describe no counterpart at OpenAI. Microsoft and Meta have their own rules: see Microsoft Ads credits for invalid clicks and Meta refunds for bot clicks.

Two more details from OpenAI's reporting docs:

  • The clicks metric is "the number of ad clicks recorded." OpenAI doesn't say whether invalid clicks are removed before that count.
  • Recent clicks "may change as processing completes," and spend reflects finalized activity that can update later. Compare days that are a few days old.

OpenAI's Advertising Terms, in the version we saw in search results, say metric-based fees are based on OpenAI's own measurement and data. OpenAI's site couldn't be read to confirm the wording.

Step 2: Tag every ad click

Add a tracking template so each landing from an ad carries its IDs. OpenAI's own example:

utm_source=openai&utm_campaign={campaign_id}&utm_content={ad_id}&click_id={oppref}

You can set it for the account, a campaign, an ad group or an ad. Your server logs then show which ad each landing came from, and click_id holds oppref, OpenAI's attribution ID, when the click supplies one. To split these visits out in GA4 as well, see how to tell ChatGPT ad clicks from organic ChatGPT traffic.

Step 3: Compare OpenAI's clicks with what reached your site

  1. Pull OpenAI's clicks by ad and day. Through the API, call Insights with aggregation_level=ad and time_granularity=daily. Reports use the ad account's time zone.
  2. Count tagged landings in your logs for the same days and time zone, grouped by utm_content.
  3. Take out OpenAI's crawler. OAI-AdsBot visits pages submitted as ads to check their safety, so its visits are page checks, not people clicking your ad. It names itself in the user agent, and OpenAI lists its IP ranges at openai.com/adsbot.json. If your firewall blocks it, see whether bot protection stops ChatGPT ads from being approved.
  4. Sort the rest into human-looking and bot-looking landings, using the signs in Step 4.

How to read it: if OpenAI's clicks roughly match all landings, bots included, you were probably billed for the bots. If they match only the human-looking ones, OpenAI probably filtered them. This is our method, not OpenAI's.

Why not GA4? Google says GA4 drops traffic from known bots and spiders automatically, and you can't turn that off or see how much it removed. A bot-heavy day can look normal there.

Step 4: Look for the signs of bot clicks

TrafficGuard, which sells click-fraud protection, looked at 40,021 ChatGPT ad clicks from 47 advertisers between August 24 and September 14, 2026. It flagged 1,592 as invalid: about 4% overall by PPC Land's math, and 0.1% to 34% depending on the advertiser. Overall, that's lower than on Google: PPC Land notes TrafficGuard's infographic put the same advertisers' overall Google Ads invalid rate at 10.8%. What stood out was the kind of invalid click. TrafficGuard says it was "highly concentrated," from a few sources on hosting and proxy networks, not scattered bots. Its signs:

  • Cloud and hosting IPs. Clicks from hosting, proxy and malicious IPs ran at 3.6 times the rate of the same advertisers' Google Ads clicks. Four networks sent 742 of the 1,592 invalid clicks.
  • Devices that can't be real. It flagged 343 clicks whose hardware didn't match the device they claimed, such as phones reporting 96 or even 192 processor cores.
  • The same sources every day. The same IP ranges or device setups came back day after day.

Two more signs from general practice, not the study: many clicks from one IP or at steady intervals, and visits with no engagement (no scrolling, no second page, no JavaScript running, or junk form fills).

PPC Land notes TrafficGuard's release doesn't say the analysis was audited externally, and that it counts clicks that reached advertisers' sites, not clicks OpenAI billed.

Two cautions for B2B. A data center IP alone doesn't prove a bot: corporate VPNs and security proxies often browse from them, so check behavior too. And OpenAI's platform breakdown files missing or unrecognized client information under unknown, along with Windows and macOS app activity, so a big unknown share is worth a look but proves nothing.

Step 5: Keep bot form fills out of a Conversions campaign

A Conversions campaign uses your conversion event, with ad quality, relevance, click likelihood and conversion likelihood, to favor clicks likely to lead to that event. If bots fill your form and you send lead_created for each one, you teach it to find more clicks like theirs. OpenAI says incomplete or incorrect tracking "can make reporting and optimization less effective."

So send the event only for leads that pass your checks, such as a real work email, from your server through the Conversions API. For the setup, see how to track conversions from ChatGPT ads.

Step 6: Limit exposure while you investigate

  • There's no IP exclusion in OpenAI's docs. Its targeting docs offer locations, platforms and custom audiences, plus context hints, so you can't stop an IP range from seeing your ads.
  • Tighten locations if bad clicks come from places you don't sell. See why ChatGPT ads bring leads from other countries.
  • Drop a platform if bad clicks cluster on one. Through the API, campaigns can target the iOS app, the Android app, and desktop, iOS and Android web browsers separately.
  • Block bad IPs on your own site to stop junk form fills. That doesn't stop OpenAI from counting the click. Keep OAI-AdsBot allowed.
  • Cap spend with the campaign budget, and with an account spending limit if your account has one.

A click-fraud tool can do the log work, but with no IP exclusion it can't keep anyone from seeing your ads. For the trade-offs, see whether click fraud tools are worth it for B2B.

Step 7: Ask OpenAI for a credit

OpenAI's developer docs describe no claim process, so write to its ads support with your numbers: ad account ID, campaign and ad IDs, dates in the account's time zone, OpenAI's clicks next to your human-looking landings, and sample log lines (time, IP, network owner, user agent, click_id).

Don't wait. We couldn't read OpenAI's help center or Advertising Terms, but in the versions seen in search results, the help center gives [email protected] for Ads Manager support and asks billing requests to include an invoice or charge ID and the date range. The terms ask you to dispute fees "within thirty days of the activity in dispute" and leave any adjustment to "OpenAI's sole discretion."

Keep expectations modest. The only public ChatGPT ads billing dispute we found wasn't about bots. The Register reported on August 13, 2026 that OpenAI support confirmed "£60.72 in invalid charges" from ads that kept running after an advertiser paused them, then declined a credit, citing its Advertising Terms.

Common problems

ProblemWhy it happensFix
GA4 shows far fewer visits than OpenAI's clicksGA4 drops known bots without saying how manyCompare with server logs
Logs show OpenAI visits that aren't clicksOAI-AdsBot checks pages submitted as adsFilter it by user agent and openai.com/adsbot.json
A Conversions campaign brings junk leadsBot form fills are sent as conversionsSend events only for leads that pass your checks
Sources (17)

Related resources