Key takeaways
- ChatGPT Ads now lets you upload raw or hashed emails and phone numbers as audience filters, mirroring Customer Match on Google and Meta.
- The feature is a first-party data and consent decision first, and a targeting upgrade second. Uploading a list your privacy notice never covered is a UK GDPR and PECR risk.
- Measurement on ChatGPT Ads is immature. Treat early results as directional and set up your own UTM tagging and analytics rather than relying on OpenAI’s reporting.
- Clean, dedupe and segment your customer file by value before hashing anything. Bad data is amplified, not fixed, by a new platform.
- Cap your initial test budget and never divert spend from proven Google or Meta campaigns into a channel you cannot yet measure properly.
ChatGPT Ads audience lists are custom audience uploads for OpenAI’s ad platform: advertisers can reportedly use emails or phone numbers to target, suppress or seed audiences. For UK PPC teams, the real question is not whether the targeting looks useful, it is whether your consent, list hygiene and measurement are strong enough before customer data leaves your CRM.
That distinction matters. A customer list is only an advantage if it is cleaner than the platform’s default targeting. Upload a stale export, or a file where permission is murky, and you have not created a smart test. You have created a governance problem with media spend attached.
We are interested in ChatGPT Ads, but we are not treating the arrival of audience uploads as a reason to rush. It is the same line we take with any early ChatGPT ads test: prove the data basis first, then decide whether the channel deserves budget.
ChatGPT Ads audience lists: what the feature appears to do
The audience upload flow is described in OpenAI’s own support material for OpenAI Custom Audiences setup. In practical terms, it looks much like the customer-list products PPC teams already know: upload customer identifiers, build an audience, then use that audience for campaign targeting or exclusions.
If that sounds familiar, it should. Google’s Google Customer Match requirements and Meta’s Meta Custom Audiences guidance follow the same broad pattern, even though the eligibility rules, match logic and reporting differ by platform. The mechanism is simple: the platform compares uploaded identifiers with logged-in users, then lets the advertiser reach or exclude the matched group.
The important caveat is that ChatGPT Ads audience lists are still early-stage. Do not assume the controls, reporting or policy detail will match Google Ads or Meta on day one. A mature interface can make a risky data upload look routine, but the compliance and measurement questions do not disappear because the button is easy to click.
The first-party data risk most teams will underprice
Most advertisers will see audience uploads as a targeting feature. We see them as a data-sharing decision first. The moment you upload a customer file to a third-party ad platform, you need to know where the data came from, what the customer agreed to, who is processing it, and how long it is retained.
Hashing helps protect identifiers in transit, but hashing does not create consent. That is the line too many marketing teams blur. If the customer never agreed to their data being used for advertising on third-party platforms, changing an email address into a hash is not a magic permission slip.
Under UK GDPR and PECR, purpose matters. If your privacy notice only mentions email marketing, CRM management or service updates, it may not cover sharing customer identifiers with an AI ad platform for targeting. The ICO direct marketing guidance is the right starting point, especially where profiling, matching or repurposing data is involved.
This is the same discipline we apply to offline conversion tracking setups. Once data leaves your owned systems and lands inside a platform, the question is no longer just, “Will it improve performance?” It is, “Can we justify this use of the data if someone asks?”
Do not assume Google-level match rates
Google Ads has years of match-rate history, advertiser documentation and conversion reporting around Customer Match. ChatGPT Ads does not yet have the same track record. That changes how much confidence you should place in any early result.
In our own account audits, measurement is still the weakest part of many mature PPC setups. Across £1.1 million of UK Google Ads spend we analysed in Q2 2026, tracking issues appeared in 56% of accounts. If advertisers still mismeasure Google Ads after two decades of platform maturity, they should expect a newer channel to be messier, not cleaner.
That does not mean do not test. It means budget for uncertainty. Early audience performance should be treated as directional until you have independent tracking, enough conversion volume and a clean comparison against existing channels.
Model CPA as a test result, not a forecast
New ad inventory often looks efficient when demand is thin. Then more advertisers arrive, auctions tighten, and the early CPA stops looking so flattering. We have seen this pattern repeatedly across paid social placements, Performance Max inventory expansion and retail media.
Launch economics are not steady-state economics. If a ChatGPT Ads audience list test produces a cheap CPA in week one, do not build next quarter’s forecast around it. Hold back budget, widen the confidence interval, and ask whether the result still stands once frequency rises and colder users enter the mix.
PPC Geeks’ View
The specific problem we expect to see is data hygiene. Teams will grab whichever customer list is easiest to export, hash it, and upload it to ChatGPT Ads audience lists because they want to be early. That file is often the same tired export recycled across Google and Meta for two years, full of duplicates, unsubscribed contacts and customers who should be suppressed, not targeted.
We see this most often in lead generation accounts where marketing owns the ad platforms but sales or operations owns the CRM. Nobody has agreed which field is the source of truth, so the list that reaches the ad account is simply the one someone could pull fastest. On a mature platform, that wastes budget. On a newer one with less familiar controls, it can also mean serving ads to people who explicitly opted out.
Every new ad platform tempts teams to upload first and check consent later. On ChatGPT Ads, that instinct is a liability, not a shortcut. Clean the list, confirm the permission, then test.
Chris S, Client Director, PPC Geeks
The takeaway is simple: do not upload a single file until someone can point to the exact consent basis for using it on a third-party AI advertising platform. This is the type of issue we look for in a free Google Ads audit, particularly where first-party data, tracking and consent are intertwined. If you want a partner to test new channels without treating governance as admin, our Google Ads agency team can help build that process properly.
A practical readiness plan before you upload
You do not need to wait for ChatGPT Ads audience lists to become a mature product before preparing your data. The work that makes them safer and more useful is the same work that improves Google Ads, Meta, Microsoft Advertising and CRM-led reporting.
- Audit the source file before you hash anything. Remove unsubscribes, deduplicate records, strip out hard bounces and exclude anyone who opted out of third-party advertising use. A clean 8,000-record file is better than a 20,000-record file where half the audience should not be there.
- Check the consent basis against your privacy notice. Look for plain wording that covers advertising, platform matching, profiling where relevant, and third-party processors. If it is not there, do not rely on wishful interpretation.
- Segment by commercial value, not file size. Separate high-value customers, recent buyers, lapsed customers, churned users and sales-qualified leads. One blended list hides the signal you are trying to buy.
- Use suppression as seriously as targeting. Excluding recent converters, existing customers or opted-out users is often the first place audience lists pay back. Suppression is not boring admin. It is budget protection.
- Cap the first test budget. Decide the amount you can afford to lose before the campaign goes live. Do not raid proven Google or Meta activity to fund a channel you cannot yet measure properly. That mistake sits near the top of our list of common PPC management mistakes.
- Track independently from day one. Use UTMs, GA4, CRM source fields and offline conversion checks where possible. Platform numbers are useful, but they should not be the only version of the truth.
How to judge early performance without fooling yourself
The first ChatGPT Ads audience test should answer one narrow question: does this audience produce enough qualified engagement to justify a larger, cleaner experiment? It should not be asked to prove the full business case for the channel.
For lead generation, judge quality as well as volume. A lower CPL is meaningless if sales rejects the leads or if demo requests come from poor-fit accounts. For ecommerce, look beyond last-click revenue and compare new customer rate, repeat purchase behaviour and margin. If you are already working through AI-era search ad fixes, use the same discipline here: clear naming, clean UTMs, defined audiences and documented assumptions.
One rule of thumb: if you cannot explain how a conversion was captured, matched and valued, you should not use it to scale spend. That is true on Google Ads, and it will be even more true on a newer platform where reporting norms are still forming.
Where ChatGPT Ads audience lists fit in your media plan
ChatGPT Ads audience lists are a signal, not a finished playbook. OpenAI is assembling pieces that look familiar to performance marketers: campaign management, audience uploads, privacy terms and conversion reporting. The opportunity is real, but so is the risk of treating a new platform like a mature one before it has earned that trust.
Our view is to prepare now and spend carefully. Get your lists clean, document consent, build exclusions, and make sure measurement does not depend entirely on OpenAI’s own reporting. The advertisers who win on new channels are rarely the ones who upload first. They are the ones whose data was ready before the auction became crowded.
For policy context, review the OpenAI EU privacy policy alongside your own privacy notice before any upload. If you are not sure whether your customer lists and PPC tracking are ready for this kind of test, that uncertainty is worth resolving before money moves. A free PPC audit is the fastest way to see where the weak points are.
Frequently asked questions
What are ChatGPT Ads audience lists?
They are a new tool inside ChatGPT Ads Manager, under the Tools section labelled Audiences, that lets advertisers upload raw or hashed emails and phone numbers to use as audience filters for campaigns. It works like Customer Match on Google Ads or Custom Audiences on Meta.
Is uploading customer data to ChatGPT Ads compliant with UK GDPR?
Only if your consent basis genuinely covers it. If your privacy notice never mentioned sharing hashed identifiers with third-party AI ad platforms, uploading is a compliance risk. Check your policy against the ICO’s direct marketing guidance before you upload anything.
Should UK advertisers move budget to ChatGPT Ads now?
No. The platform’s measurement is immature, so cap any test with a small fixed budget you can afford to lose. Do not divert spend from proven Google or Meta campaigns to fund a channel you cannot yet track reliably.
How do I make my customer list ready for ChatGPT Ads?
Clean the file first: remove unsubscribes, dedupe records, and strip anyone who opted out of third-party sharing. Then segment by value so you can build lookalikes from high-value customers and suppress churned ones. Only hash and upload after that.
Can I trust ChatGPT Ads reporting for ROAS decisions?
Not on its own yet. Set up independent conversion tracking with UTM tagging and your own analytics so you can judge performance against a source you control. Early match rates and attribution on a new platform are far less transparent than Google’s.
