To track ChatGPT ads, add UTM parameters to every landing URL. OpenAI’s own examples use utm_source=openai&utm_medium=paid. Put them in the Ads Manager tracking template so the {campaign_id} and {ad_id} macros fill in on each click. UTMs only label visits. To count purchases and leads you also need the OpenAI Pixel and/or the Conversions API. Both rely on the oppref click reference that OpenAI appends to every ad click.
If you only want the finished link, the free ChatGPT ads UTM builder gives you both a full tagged URL and a ready-to-paste Ads Manager template with the macros.
How ChatGPT ads tracking fits together
A ChatGPT ad click can carry three different kinds of tracking data. Each one has its own job, so keep them apart when something looks wrong.
| Layer | What it is | Who adds it | What it answers |
|---|---|---|---|
| UTM parameters | utm_source, utm_medium, utm_campaign and so on |
You, as static values or through the tracking template | Which campaign and ad brought this session, in GA4 or any analytics tool |
| Macros | {campaign_id}, {ad_id}, {oppref} inside the tracking template |
You place them; OpenAI fills them in on each click | Exact IDs that join analytics rows to Ads Manager rows |
oppref click reference |
A parameter OpenAI appends to the landing URL | OpenAI, automatically | Which ad click a later conversion belongs to, for the Pixel and Conversions API |
Don’t count on a referrer header such as chatgpt.com showing up on ad clicks. That behaviour isn’t documented, so treat UTMs and oppref as the reliable signals.
Choose a UTM convention and keep it
OpenAI’s documentation uses utm_source=openai and utm_medium=paid in its examples. Many advertisers and tools use utm_source=chatgpt and utm_medium=cpc instead. Both work. What matters is that every campaign, ad and teammate uses the same values. UTM values are case-sensitive, so ChatGPT and chatgpt become separate sources in your reports.
A practical recommendation:
| Parameter | Recommended value | Why |
|---|---|---|
utm_source |
openai |
Matches OpenAI’s examples. It also covers future placements that aren’t strictly “ChatGPT”. |
utm_medium |
paid |
Marks the traffic as paid. GA4 generally puts mediums like paid or cpc into a paid channel rather than referral or organic. Check where the sessions actually land in your Traffic acquisition report. |
utm_campaign |
se-office-cleaning-2026q4 |
Readable and lowercase. Ideally it matches the campaign name in Ads Manager. |
utm_id |
{campaign_id} |
Filled in by OpenAI, so it joins exactly to Ads Manager data. |
utm_content |
{ad_id} or a readable ad label |
Tells ads apart without maintaining a manual lookup. |
utm_term |
Optional, e.g. emergency-leak |
Useful for the context-hint theme behind an ad group. Leave it empty rather than guessing. |
If you already have months of utm_source=chatgpt data, you don’t have to switch. Pick one convention, write it in your naming sheet, and only change it at a clean break, such as the start of a quarter, with a note in your reporting. Mixing conventions mid-campaign does more harm than either choice.
Never put personal data such as names, emails or customer IDs in a tracking URL.
Use the tracking template and its macros
In Ads Manager you can set tracking parameters as a query-string template. In the API the field is called query_string_template. Static values persist on every click. The template also accepts three macros that OpenAI replaces at click time:
| Macro | Replaced with | Typical use |
|---|---|---|
{campaign_id} |
The campaign’s ID | utm_id={campaign_id} for exact joins |
{ad_id} |
The ad’s ID | utm_content={ad_id} to compare ads |
{oppref} |
The click reference | Passing the reference under your own parameter name, for example to a CRM or booking system that only stores specific fields |
A template that follows OpenAI’s convention looks like this:
utm_source=openai&utm_medium=paid&utm_campaign=se-office-cleaning-2026q4&utm_id={campaign_id}&utm_content={ad_id}
The final landing URL on a click then looks roughly like this. OpenAI fills in the IDs and appends oppref itself:
https://example.com/office-cleaning/?utm_source=openai&utm_medium=paid&utm_campaign=se-office-cleaning-2026q4&utm_id=<campaign id>&utm_content=<ad id>&oppref=<click reference>
Keep the macros literal, with braces and without URL encoding. A template that contains %7Bcampaign_id%7D won’t be replaced.
Which level wins
OpenAI’s documented precedence, from lowest to highest, is:
Feed < Account < Campaign < Ad group < Ad
The most specific level that has a template takes precedence. The classic mistake is a careful account-level template silently overridden by an old ad-level one. For most small advertisers it’s simplest to use one level, usually campaign, and leave the others empty.
Keep the full URL under 2,048 characters
OpenAI caps landing URLs at 2,048 characters. Long product URLs with filters plus UTMs, macros and oppref can get close, so keep values short.
Mid-setup shortcut: the UTM builder normalises values to lowercase with hyphens, warns above 2,048 characters, moves #anchors after the parameters, and outputs the matching Ads Manager template with {campaign_id} and {ad_id} already in place.
Protect the oppref click reference
OpenAI appends oppref to the landing URL on every ad click. The OpenAI Pixel reads it and stores it in a first-party cookie. When a conversion fires later, even on another page, it can be tied back to the click. Lose oppref and the click can no longer be attributed in ChatGPT reporting. The visit itself still arrives, which makes the problem hard to spot.
Common ways oppref disappears:
- Redirects that drop the query string. Examples are http to https, www to non-www, language redirects, or a trailing-slash rule that rebuilds the URL.
- Single-page apps whose router rewrites the URL before the Pixel has loaded.
- “Clean URL” plugins or CDN rules that strip unknown parameters for caching.
- Consent banners that reload the page without the original parameters.
Test by opening the final tagged URL with a dummy value, such as &oppref=test123. Then check the address bar after every redirect, and confirm that the cookie is set once consent is given. The landing-page guide covers the rest of the destination checks.
Expect OAI-AdsBot review visits
Before and while ads run, OpenAI’s ad-review crawler, OAI-AdsBot, visits your landing pages. It visits them with the tracking parameters. In analytics those visits can look like real paid sessions from openai / paid. They aren’t billed clicks, and they will never convert.
What to do about it:
- Expect a few “paid” sessions before your first real impression, and after you edit an ad.
- Where your analytics or server logs allow it, exclude the OAI-AdsBot user agent from paid reporting.
- Don’t block the crawler in robots.txt or your firewall. Ad review needs to reach the page.
Set up conversion tracking: Pixel and Conversions API
UTMs tell GA4 where a visit came from. They don’t tell ChatGPT reporting that a sale happened. For that, OpenAI offers two integrations, and you can run them together:
| Method | How it works | Strengths | Watch out for |
|---|---|---|---|
| OpenAI Pixel | JavaScript on your site. It stores oppref in a first-party cookie and sends events from the browser. |
Quick to install and reads oppref automatically |
Affected by consent choices, ad blockers and browser limits |
| Conversions API (CAPI) | Your server sends events to OpenAI | Survives browser limits. It’s also the only way to send app_installed and app_opened. |
Needs development work, and you must pass along the click reference |
If you send the same conversion from both, give it the same event ID in each. OpenAI uses that shared ID to deduplicate, so one purchase doesn’t count twice. Optional extras include automatic advanced matching, which hashes data in the browser, and modeled conversions. Check how both fit your privacy notice before switching them on.
Supported events
OpenAI’s supported standard events are:
| Event | Typical use |
|---|---|
page_viewed |
Any page view |
contents_viewed |
Product or content detail view |
items_added |
Add to cart |
checkout_started |
Checkout begins |
order_created |
Completed purchase |
lead_created |
Form or quote request submitted |
appointment_scheduled |
Booking confirmed |
registration_completed |
Account or sign-up completed |
subscription_created |
Paid subscription started |
trial_started |
Free trial started |
custom |
Anything else you define |
app_installed, app_opened |
App events, Conversions API only |
Conversion-optimised campaigns can only optimise towards standard events, not custom. Map your most important outcome to the closest standard event, such as lead_created for a quote form, even if you also send a custom event. Remember that the Conversions objective is still billed per click or per thousand impressions, never per conversion.
Attribution defaults
ChatGPT reporting attributes conversions with a default 30-day click window and 1-day view window. That is a reporting setting, not a technical limit on the cookie. GA4 uses its own attribution model and lookback windows, which is one reason the two systems disagree. Write down which windows you compare, every time.
Consent, GDPR and what you’ll lose
In Sweden and the rest of the EU, the Pixel’s cookie and browser events generally need the visitor’s consent before they fire. If your consent tool blocks them until the visitor accepts, people who decline won’t appear in Pixel-based ChatGPT reporting. They often won’t appear in GA4 either, depending on your consent mode setup. A server-side Conversions API event also needs a legal basis and has to match what your privacy notice says. It isn’t a way around consent. This isn’t legal advice, so check your setup with whoever owns privacy at your company.
In practice:
- Expect both ChatGPT reporting and GA4 to under-count compared with your order system. How much depends on your consent rate.
- Make sure the consent banner doesn’t reload the page and strip
opprefand the UTMs. - Note that in the EEA and Switzerland, ads are matched on context, with no ads personalization or custom audiences. That affects targeting, not your obligation to ask for consent.
Step-by-step setup
- Write the naming sheet. Fix
utm_source,utm_mediumand theutm_campaignpattern. Add one row per campaign. - Build the template. Use the UTM builder, or write
utm_source=openai&utm_medium=paid&utm_campaign=…&utm_id={campaign_id}&utm_content={ad_id}. - Choose one level. Paste the template at campaign level, unless you have a reason to go higher or lower. Clear any older templates at other levels.
- Install the Pixel. Load it on every page through your consent tool, and confirm it stores
opprefafter consent. - Pick and map events. Choose the standard event that matches your main outcome. Add
checkout_startedoritems_addedas supporting signals if you sell online. - Add the Conversions API if you can. Send the same events from the server with shared event IDs.
- Test the redirect chain. Open the final URL with
&oppref=test123on mobile and desktop, and check that all parameters survive. - Test a conversion. Make a test enquiry or an approved test order, label it as a test in your own system, and check that the event arrives.
- Check GA4. Confirm the session appears under your chosen source and medium in real-time and Traffic acquisition.
- Run the pre-flight checklist before you submit the campaign for review.
Reconcile ChatGPT reporting with GA4
The two systems count different things, so they never match. Ads Manager reports impressions, clicks, spend, CTR, average CPC, average CPM and conversions. The API adds CPA, post-click conversion rate, attributed sales and ROAS. GA4 counts sessions and its own key events.
| Why the numbers differ | Direction | What to do |
|---|---|---|
| A click isn’t a session (bounces before the tag loads, declined consent) | Sessions lower than clicks | Compare trends, not absolute values |
| OAI-AdsBot review visits | Sessions slightly higher | Filter by user agent where possible |
| Different attribution windows and models | Either way | Write down both windows when comparing |
| Click data refreshes about every 15 minutes, while spend can lag 7–8 hours | Today’s numbers look off | Compare completed days only |
| Time zones differ between the ad account and GA4 | Day totals shift | Compare weeks, or align time zones |
oppref stripped by a redirect |
ChatGPT conversions too low | Fix the redirect chain |
A worked example with invented numbers: 80 clicks in Ads Manager, 66 openai / paid sessions in GA4, 5 lead_created conversions in ChatGPT reporting and 4 form leads in GA4. That doesn’t prove 14 clicks “disappeared”. Declined consent, review visits and different attribution windows explain gaps like this. Check redirects, time zones and date ranges, then judge the campaign on the leads your sales team qualified. If visits are showing up as direct traffic, trace where the campaign context is lost. If conversions are double-counted, review your event deduplication.
Troubleshooting table
| Symptom | Likely cause | Fix |
|---|---|---|
Sessions show up as (direct) / (none) |
A redirect or SPA router drops the query string | Preserve the query string on every redirect, and load analytics before routing |
utm_id shows the literal {campaign_id} |
Macro URL-encoded, or used in a manually typed URL rather than in the template | Put the macro in the tracking template, unencoded |
Source split across ChatGPT, chatgpt, openai |
No shared convention | Fix one convention and lowercase everything |
| “Paid” sessions before launch | OAI-AdsBot review visits | Expected. Filter by user agent. |
| Ads Manager shows clicks but zero conversions | Pixel blocked by consent, oppref stripped, or wrong event mapped |
Test with &oppref=test123, check the cookie after consent, and check the event name |
| Conversions double-counted | Pixel and CAPI send without a shared event ID | Send the same event ID from both |
| Template changes have no effect | An ad- or ad-group-level template overrides it | Clear the more specific levels |
| Landing URL rejected | Longer than 2,048 characters | Shorten the UTM values, and drop filter parameters you don’t need |
| Campaign can’t optimise for your event | It’s a custom event |
Map the outcome to a standard event |
Frequently asked questions
Should utm_source be openai or chatgpt?
Either works, as long as you use it everywhere. OpenAI’s own examples use utm_source=openai&utm_medium=paid, so that’s a sensible default for new accounts. If you already have history under chatgpt, keep it and avoid mixing the two.
Does ChatGPT add UTM parameters automatically?
No. OpenAI appends its own click reference, oppref, to ad clicks, but UTMs are up to you. Add them as static values or through the tracking template, where {campaign_id} and {ad_id} fill in automatically.
What is oppref and can I remove it?
oppref is the click reference OpenAI adds to the landing URL. The Pixel stores it so that later conversions can be tied to the ad click. Don’t strip it in redirects, routing or URL-cleaning plugins, or ChatGPT conversion reporting will miss those conversions.
Why do I see paid ChatGPT sessions before my ads went live?
These are most likely OAI-AdsBot, OpenAI’s ad-review crawler. It visits landing pages with your tracking parameters attached. The visits aren’t billed and won’t convert, so filter them out where your tools allow.
How does ChatGPT ads conversion tracking handle duplicates?
If you send the same event from the Pixel and the Conversions API, give both the same event ID. OpenAI deduplicates on that ID, so one purchase or lead is counted once.
What attribution window does ChatGPT ads use?
The default is a 30-day click window and a 1-day view window. GA4 uses its own model, so compare the two over the same completed date range, and expect some difference.
How AthillyAds helps
AthillyAds builds ChatGPT ad campaigns from your website. It writes the ad copy and context hints, runs pre-flight checks before launch, and puts consistent tracking on every ad before export or publishing. Connected reporting then shows clicks, spend, CTR and CPC per campaign, next to your own analytics. To see how your business would be matched, try the free context hints generator. The free plan covers one full campaign, and pricing explains Pro. Ad spend is always paid separately to OpenAI. Once live, follow the first-week review guide.
Sources and scope
- OpenAI Help: Ads in ChatGPT, the basicsRead
- OpenAI Help: Conversion measurementRead
- OpenAI Help: Conversion-optimized campaignsRead
- OpenAI Ads developer docs (campaign management, reporting, changelog)Read
- OpenAI Ads developer docs: supported eventsRead
- Google Analytics Help: URL builders and campaign parametersRead
Working methods and examples are editorial suggestions. Check current platform requirements and available features before implementation.
