ChatGPT Ads custom audiences require at least 25,000 matched users before an audience can run, and OpenAI recommends 100,000. That floor is roughly 25 times Meta's practical delivery threshold, 83 times LinkedIn's minimum, and 250 times Google Customer Match's. For most B2B advertisers, whose usable first-party lists rarely clear that bar once match rates are accounted for, customer-list targeting on ChatGPT Ads is effectively an enterprise feature at launch.
That is not a reason to sit out ChatGPT Ads. It is a reason to know which lever to pull. The channel was built contextual-first, so the targeting engine that matters for B2B is still context hints, and the audience control any advertiser can use today is location exclusion. Here is what the 25,000 number actually means, how it compares, and what to do instead.
- The floor: 25,000 matched users, 100,000 recommended
- Why it bites: only matched users count, so your raw list needs to be much larger
- The comparison: 25 to 250 times higher than Google, Meta, and LinkedIn
- What still works for B2B: context hints (no size floor) and location exclusion
- Who it is for: enterprise and large-user-base brands, not most mid-market B2B
What ChatGPT Ads custom audiences are
Custom audiences let you upload a list of customer emails or phone numbers, raw or SHA-256 hashed, so ChatGPT Ads can match them against its logged-in users and let you target or suppress that specific group across your campaigns. It is the same first-party-data primitive that Meta introduced in 2012 and Google shipped as Customer Match in 2015, and OpenAI added it to ChatGPT Ads on May 14, 2026. The full mechanics, from hashing to include-versus-exclude, live in the custom audiences walkthrough and the glossary definition.
This post is about one number inside that feature that decides who can actually use it. OpenAI's documentation now states that each custom audience must include at least 25,000 matched users before it can run. That single line quietly puts customer-list targeting out of reach for a large share of the B2B advertisers the platform just opened its doors to.
The 25,000 minimum, and why “matched” is the word that matters
Read the requirement carefully: the floor is 25,000 matched users, not 25,000 uploaded rows. When you upload a list, OpenAI normalizes it, hashes any plaintext, and matches it against ChatGPT's logged-in user graph. OpenAI's own documentation notes that invalid values, duplicate values, and identifiers it cannot match do not count toward the audience size. Only the successful matches do.
That distinction is the whole story. You do not need a list of 25,000. You need a list large enough that, after ChatGPT discards everyone it cannot find, 25,000 remain. On a mature platform like Meta or Google, where nearly everyone has an account, match rates on a clean list run high. On ChatGPT, the population that matters is smaller: only free and lower-tier users see ads, and for a B2B contact list the overlap with regular ChatGPT users is thinner still. The lower your match rate, the bigger the raw list you have to start with.
The 25,000 is not the size of your list. It is the size of your list after ChatGPT throws away everyone it cannot find.
OpenAI recommends audiences of at least 100,000 users, four times the floor, which is a strong hint that even at 25,000 an audience performs thinly. For context on the upload itself, the platform accepts a CSV or TXT file up to 500 MB holding up to 5 million identifiers, one identifier type per file, and processing takes about 20 to 30 minutes. None of those limits is the constraint that bites. The 25,000-match floor is.
How the floor compares to Google, Meta, and LinkedIn
The reason 25,000 lands so hard is that every other platform a B2B marketer already uses set its customer-list floor an order of magnitude or two lower, and most of them have been lowering it, not raising it.
Google cut its Customer Match minimum from 1,000 to 100 users for Search campaigns in 2025. LinkedIn Matched Audiences start at 300 member accounts. Meta will technically create a custom audience at 100 people, though in practice you need around 1,000 for reliable delivery. Against that backdrop, ChatGPT Ads asking for 25,000 matched users is not a slightly higher bar; it is a different category of requirement.
There is a difference in kind, too, not just degree. Google's and Meta's low numbers are practical delivery thresholds, the point below which the auction cannot find enough people to serve. OpenAI's 25,000 is a hard, published rule that blocks the audience from being used at all. You can build the audience, watch it match, and still be locked out until it clears the line.
Why this shuts out most B2B
Put the match math and the comparison together and the consequence for B2B is direct. A mid-market B2B company often has a CRM in the single-digit thousands of active contacts. A well-run demand-generation program might have 20,000 or 30,000 known names, a meaningful share of them stale, unengaged, or working from an email the person no longer uses. That is the list before it meets ChatGPT's user graph.
Now match it. Your buyers skew toward work email addresses and toward the paid ChatGPT tiers that do not see ads, both of which pull the match rate down. A list that looked comfortably above 25,000 on paper can land well under it once matched. The company did nothing wrong; the floor is simply calibrated for consumer-scale lists, and most B2B lists are not consumer-scale.
The 25,000-match floor is calibrated for consumer-scale lists. Most B2B lists are not consumer-scale, which is exactly why the feature reads as enterprise-only at launch.
This is less a flaw than a signal about who the audience layer is for right now. It also reinforces something true about ChatGPT Ads from the start: the platform was designed contextual-first, and for the median B2B advertiser it is going to stay that way for a while. Which is the good news, because the contextual lever is the one you can use without any list at all.
What actually works for B2B instead
If a 25,000-match audience is out of reach, two levers still are not, and together they cover most of what a B2B advertiser actually needs from ChatGPT Ads today.
Context hints stay the primary targeting engine
Context hints target the conversation rather than a known list of people, and they carry no audience-size floor. You describe the topics, intent, and moments your ad should be eligible for, and the engine matches your message to relevant conversations. For a company that cannot assemble a 25,000-person matched audience, this is not a fallback; it is the main event. The craft is in writing hints that are specific enough to reach buyers and broad enough to deliver, which is covered in the guide to context hints and the five patterns.
Location exclusion is the audience control you can use today
Alongside custom audiences, OpenAI added location and audience exclusion controls in July 2026. Location exclusion is the one that matters for advertisers under the floor: it lets you stop a campaign from serving in specific states, designated market areas, or ZIP codes, and it has no audience-size requirement. Use it to cut regions you do not sell into, suppress spend around markets you cannot support, or keep a campaign off geographies that inflate cost without producing pipeline. It pairs naturally with the state, DMA, and ZIP geo-targeting controls that shipped in May.
One honest caveat on the other kind of exclusion. Suppressing a custom audience, for example to stop advertising to existing customers, is appealing, but a suppression list is a custom audience like any other. On the plain reading of OpenAI's rule, it is subject to the same 25,000-matched-user floor, so a small customer list cannot be used to suppress any more than it can be used to target. If customer suppression is central to your plan, confirm the behavior in your own account rather than assuming the floor is waived for exclusion.
Who custom audiences is right for
The 25,000 floor is not a problem for everyone. It is a clean fit for a specific set of advertisers, and it is worth naming them honestly so you can tell which side of the line you are on.
- Enterprise B2B with large, clean databases. If your marketable list runs into the hundreds of thousands, you will clear 25,000 matched even at a modest match rate, and custom audiences give you real retargeting and account-list activation.
- Product-led and freemium companies. A large base of free-tier or trial users is exactly the kind of consumer-scale list the floor was built for, and it is often your best seed for re-engagement.
- Large-list B2C and ecommerce. Retailers and consumer brands with sizable customer files are the native audience for this feature, the same profile that made Custom Audiences work on Meta.
- Not most SMB and mid-market B2B. If your usable list is in the low tens of thousands or smaller, plan around context hints and location exclusion, and treat custom audiences as something to grow into.
How to prepare if you are close to the threshold
If you are within striking distance of 25,000 matched, a few moves improve your odds of clearing it without inflating the file with junk that will not match anyway.
- Consolidate before you upload. Merge your marketing, product, and sales lists into one deduplicated file. Duplicates do not count toward the matched total, so cleaning them out first gives you an accurate read on where you stand.
- Prefer your freshest, most engaged records. Recently active contacts match at higher rates than a decade of accumulated names. A smaller, cleaner list can out-match a larger, staler one.
- Use first-party data only. OpenAI's June 2026 Ad Tools Terms prohibit purchased or broker-sourced lists. Padding a list with bought data to reach the floor is both against the rules and unlikely to match.
- Hash correctly. Normalize before you hash: lowercase and trim emails, put phone numbers in E.164 format, then apply SHA-256. The custom audiences walkthrough shows the exact hashing step.
Whether or not you clear the floor, install the conversion pixel now. It is the second identity primitive on the platform, it works on the contextual campaigns you are already running, and it is what turns list-building patience into measurable pipeline once your audience is usable.
Where AI-Advisors fits
Because the 25,000 floor keeps most B2B advertisers on contextual targeting, the highest-leverage work is not fighting for a customer-list audience you cannot fill. It is running context-hint campaigns well and proving they pay back, and that is where AI-Advisors sits.
- Turn Google Ads into ChatGPT context hints. The free converter drafts context hints and headlines from your highest-spend keywords, so you can stand up contextual campaigns without a matched audience.
- Measure which clicks move pipeline. The pixel install plus consistent UTM tagging keeps ChatGPT Ads clicks classified as paid traffic in GA4, so you can read cost per pipeline instead of guessing.
- Compound paid with organic. The AI Ads platform overlays paid spend with your organic AI citation data, so you can see where contextual ads and organic AI visibility reinforce each other on the same buyers.
The 25,000-user floor keeps most B2B advertisers on contextual targeting, which means the leverage is in writing context hints that land and knowing which clicks move pipeline. The AI-Advisors ChatGPT Ads integration installs the OpenAI conversion pixel in two minutes and overlays paid spend with your organic AI citation data. Free on all plans.
Try the AI-Advisors ChatGPT Ads integration →Custom audiences will get easier to reach as ChatGPT's ad-supported user base grows and, likely, as the roadmap adds lookalike expansion. Until then, the winning B2B play on ChatGPT Ads is the one the platform was built for: contextual reach, disciplined exclusion, and measurement that connects the spend to the pipeline. The 25,000 floor is not a wall around the channel. It is a wall around one feature, and it points you straight at the levers that already work.
Frequently Asked Questions
#What is the minimum audience size for ChatGPT Ads custom audiences?
Each ChatGPT Ads custom audience must include at least 25,000 matched users before it can run in a campaign, and OpenAI recommends at least 100,000. The 25,000 is a hard floor: below it, the audience is created but cannot be used. It counts matched users, not the number of rows you upload, so invalid, duplicate, and unmatched identifiers do not count toward the total.
#Why won't my ChatGPT Ads custom audience run?
The most common reason is that the audience has fewer than 25,000 matched users. When you upload a list, OpenAI hashes and matches it against ChatGPT's logged-in users, and only the successful matches count. If your uploaded list is small, stale, or skews toward people who do not use ChatGPT on an ad-supported tier, the matched count can fall well below the 25,000 floor even when the file itself looks large.
#How many contacts do I need to reach 25,000 matched users?
More than 25,000, because only matched users count. If your list matches at 60 percent you would need roughly 42,000 contacts; at 45 percent, about 56,000; at 30 percent, over 83,000. OpenAI does not publish ChatGPT match rates, so treat these as directional. The practical takeaway is that you should plan for a raw first-party list comfortably larger than 25,000, not exactly 25,000.
#Can B2B companies use ChatGPT Ads custom audiences?
Large enterprises and product-led companies with big user bases can. Most small and mid-market B2B companies cannot yet, because their usable first-party lists rarely produce 25,000 matched users once you account for match rates. For those advertisers, the working levers on ChatGPT Ads are context hints, which carry no audience-size floor, and location exclusion.
#Does ChatGPT Ads have negative targeting or audience exclusion?
Yes, in two forms. Location exclusion lets you stop a campaign from serving in specific states, designated market areas, or ZIP codes, and it has no audience-size floor, so any advertiser can use it today. Audience exclusion, which suppresses a custom audience, is a custom audience like any other, so on the plain reading of OpenAI's rule it is also subject to the 25,000-matched-user floor. Confirm the exact behavior in your own account before relying on it.
#Is 25,000 users a lot compared to other ad platforms?
Yes. Google lowered its Customer Match minimum to 100 users for Search campaigns in 2025, LinkedIn Matched Audiences start at 300 members, and Meta custom audiences generally need about 1,000 people to deliver reliably. ChatGPT Ads' 25,000-matched-user floor is roughly 25 to 250 times higher, and unlike some of those thresholds it is a hard published minimum rather than a practical recommendation.
#Why is ChatGPT Ads' custom audience minimum so high?
OpenAI has not published a reason. The most likely explanation is privacy: a high matched-size floor makes it harder to single out or confirm the presence of any one person in an uploaded list, which is the core privacy risk of customer-list matching. A larger required audience also reduces the odds that ad delivery reveals who is on a list. Whatever the rationale, the effect for advertisers is the same: small lists cannot be used.
#What should I do on ChatGPT Ads if my list is too small?
Lead with context hints, which target the conversation rather than a known user list and have no size floor, and use location exclusion to cut geographies you do not serve. Keep building a clean, consolidated first-party list so you are ready when your matched count clears 25,000, and install the conversion pixel now so you can measure the contextual campaigns you are running today.
