A technology partner logo is not a matched audience.
On September 28, 2026, LiveRamp published a short blog under Travis Clinger’s byline saying the company is now a technology partner of ChatGPT Ads. Marketers, the post says, can use RampID to activate first-party data as Custom Audiences inside ChatGPT Ads. Eligible inputs are framed broadly: CRM, loyalty, behavioral, web, app, or other first-party data. The resulting audiences can sharpen targeting, extend reach, suppress people you do not want to see the ad, and support measurement and optimization. The integration is available in 11 major markets, with a promise to expand over time to every market where ChatGPT Ads run (LiveRamp Blog, September 28, 2026).
That sentence is useful. It is also incomplete in the ways that matter to a media ops lead. The post does not name the 11 markets. It does not publish match rates. It does not include a statement from OpenAI. It does not say whether OpenAI now accepts a RampID string as a first-class audience identifier, or whether LiveRamp resolves RampID back into email, phone, hashed identifiers, or Google Advertising ID before the file lands in Ads Manager. PPC Land, ContentGrip, and Pivot News all filed independent readouts that rest on the same primary post and the same gaps (PPC Land, September 28, 2026; ContentGrip, September 29, 2026; Pivot News, September 29, 2026).
The operator job this week is not celebrating “CRM audiences on ChatGPT.” It is auditing which key actually crosses into OpenAI, which geos are live for your seat, how inclusion versus exclusion floors apply, and what measurement claim you are allowed to make. Treat LiveRamp as an identity-routing ticket until those four answers exist in writing for your account.
What LiveRamp actually claimed
Keep the claim small and exact.
LiveRamp says it is a ChatGPT Ads technology partner. It says marketers can use RampID to activate first-party audiences as Custom Audiences on ChatGPT. It says the route covers CRM and adjacent first-party sources for targeting, reach, suppression, and measurement. It says the integration is live in 11 major markets and will expand toward full ChatGPT Ads coverage over time. The only named implementer is Allegiance Group & Pursuant (AGP), working for an unnamed nonprofit client. Megan Morris, AGP’s director of integrated media solutions, is quoted on omnichannel impact and reach. LiveRamp says it expects to share results from other marketers soon. There are no figures in the post on match rates, pricing, lift, or donor response (LiveRamp Blog, September 28, 2026).
Two wording traps sit inside that short text. The headline and early lines use OpenAI’s product name, Custom Audiences. A later sentence slips into “Customer Audiences.” OpenAI’s documented feature is Custom Audiences. Treat the second wording as a vendor slip, not a second product (PPC Land, September 28, 2026). Separately, LiveRamp opens with a monthly user figure for ChatGPT. OpenAI has publicly framed scale in weekly active users at different points this year. Do not paste the vendor opener into a client deck as if it were OpenAI’s addressable ad pool. Ads still run on Free and Go surfaces with plan, age, and sensitive-topic exclusions. The addressable fraction behind any Custom Audience is smaller than the headline, and that fraction is not published in LiveRamp’s post.
This is also LiveRamp’s second ChatGPT Ads wire in four months. On June 10, 2026, the company connected its Conversions API Hub so conversion events could move server-to-server into ChatGPT ads with RampID in transit. September 28 runs the other direction: audiences into campaigns before they run. Measurement plumbing and audience plumbing are related. They are not the same ticket (PPC Land, September 28, 2026; ContentGrip, September 29, 2026).
OpenAI’s match rules still own the floor
Whatever LiveRamp routes, OpenAI’s Custom Audiences rules still decide whether the list can include, exclude, or bid-adjust.
OpenAI’s Help Center documentation (accessed September 29, 2026) says custom audiences are built from email addresses, phone numbers, hashed email addresses, hashed phone numbers, or Google Advertising IDs. You can mix supported identifier types in a CSV upload. After processing, audiences can be used for campaign-level inclusion, campaign-level exclusion, or ad-group bid adjustments. Inclusion audiences and audience bid adjustments require at least 25,000 matched users. Audiences below that floor can still be used for exclusions. OpenAI recommends working with at least 100,000 users for inclusion targeting or bid adjustments. Ads Manager does not show individual matched users. For campaigns targeting the EEA or Switzerland, OpenAI tells advertisers to avoid custom audiences while ads personalization is unavailable there.
PPC Land’s timeline matters for operators who last read the docs in July. Custom audience uploads first appeared in a gated Ads Manager rollout around May 14, 2026. July documentation set a 25,000 matched-user floor and tight file constraints. In the week beginning August 31, OpenAI removed the 25,000 floor for exclusion audiences, allowed member add/remove/replace without rebuilding, permitted mixed identifier types, lifted the old five-million member ceiling, and added GAID. The inclusion floor stayed at 25,000 (PPC Land, September 28, 2026; OpenAI Help Center, accessed September 29, 2026).
That asymmetry is load-bearing for LiveRamp’s pitch. A suppression list of a few thousand recent purchasers can now exclude without clearing 25,000. A targeting list still cannot. If your CRM segment matches at 40 percent into ChatGPT accounts, a 50,000-row file can land under the inclusion floor. Neither LiveRamp nor OpenAI has published the match rates that would tell you how often that happens in the 11 markets.
The key that crosses is the product
Here the public record stops, and this is the part agencies will invent if you do not force the question.
OpenAI’s published identifier list is email, phone, hashed variants, and GAID. RampID is not on that list. LiveRamp’s post does not say whether OpenAI now ingests a RampID carrying an OpenAI-specific partner encoding, or whether LiveRamp resolves RampID into one of the keys OpenAI already accepts before the audience arrives in Ads Manager (PPC Land, September 28, 2026; Pivot News, September 29, 2026).
Those two architectures produce different match curves, different failure modes, and different answers when a privacy review asks which identifier left your wall. If the path is resolve-to-email-or-phone, you are buying operational convenience on top of OpenAI’s existing match. If the path is native RampID ingest, you are buying a new identifier relationship that OpenAI has not documented on the Help Center page buyers actually use. Until LiveRamp or OpenAI states which key crosses, every deck that says “RampID Custom Audiences in ChatGPT” is selling routing, not a verified match protocol.
Do not confuse this with HubSpot’s September CRM partnership framing or Shopify’s catalog sync into ChatGPT Ads. Those are different pipes with different operator jobs. LiveRamp is identity collaboration infrastructure. It is not a CRM console and it is not a product feed. Briefing “we can push CRM audiences into ChatGPT like HubSpot” collapses three products into one Slack thread and guarantees the wrong owner picks up the work.
Eleven markets is not sixty-three
Geography is where the announcement is most precise and least usable.
LiveRamp says 11 major markets. OpenAI’s Ads Manager availability page, after the September 23 expansion into seven Asian markets, listed 63 countries where ads can be bought. Eleven is roughly 17 percent of that list. PPC Land notes that OpenAI’s public rollout timeline never paused at a clean total of 11, so you cannot reverse-engineer LiveRamp’s list from OpenAI’s country announcements (PPC Land, September 28, 2026; Pivot News, September 29, 2026; ContentGrip, September 29, 2026).
Europe is the sharpest unknown. When OpenAI told European users ads were coming, early personalization was limited, with selection tied to conversation topic, general location, and device type, and fuller personalization deferred behind opt-in mechanics. If any EEA market sits inside LiveRamp’s 11, a first-party audience integration collides with a platform that may still constrain personalized delivery. LiveRamp’s post does not engage that question. ContentGrip separately flags OpenAI’s guidance to avoid custom audiences for EEA or Switzerland campaigns while personalization is unavailable. Confirm your market list in-product. Do not infer coverage from ChatGPT Ads’ wider footprint or from LiveRamp’s “major markets” adjective.
What operators get wrong
Mistake one: treating the partner badge as proof of match quality. Technology partner is a commercial relationship. Match rate is an engineering and policy outcome. Demand the key that crosses and a sample match report for your file types before you promise inclusion targeting.
Mistake two: assuming your geo is one of the eleven. The markets are unnamed. Ask LiveRamp and check Ads Manager. If your priority countries are outside the list, you still have manual Custom Audience uploads under OpenAI’s rules where those markets support them. Do not brief global CRM activation.
Mistake three: briefing this as HubSpot CRM partnership or Shopify catalog. Different objects. HubSpot was sold as CRM-native ads workflow. Shopify was catalog and commerce plumbing. LiveRamp is RampID routing into Custom Audiences. Wrong analogy produces the wrong RACI.
Mistake four: ignoring the inclusion floor while celebrating suppression. Exclusion got easier after late August. Inclusion still needs 25,000 matched users. A nonprofit donor file that clears exclusion can still fail inclusion. Plan both uses separately.
Mistake five: equating audience eligibility with conversation control. Custom Audiences gate who is eligible. They do not hand you the prompts people typed, and they do not let you pick the exact turn an ad appears. ContentGrip’s read is right: eligibility and conversational relevance remain a hybrid. Do not sell deterministic prompt targeting.
Mistake six: shipping a measurement claim from a vendor “soon.” AGP is implementing. Results are promised later. Until independent or at least numbered results exist, your measurement story is “we can route the same identity layer into ChatGPT Ads,” not “RampID lifts ChatGPT ROAS.”
What media and agency ops do this week
- Open the key question in writing. Ask LiveRamp which identifier OpenAI receives: native RampID, email, phone, hashed variants, GAID, or a mix. Ask whether partner-encoded RampIDs are translated at the boundary. Save the answer next to OpenAI’s Help Center identifier list.
- Inventory markets for your seat. Get the named 11 from LiveRamp. Cross-check against Ads Manager availability and against any EEA or Switzerland personalization limits. Mark each priority country as live, not live, or live-but-contextual-only.
- Split inclusion and exclusion playbooks. Build exclusion tests on high-certainty recent converters even if the matched count is small. Build inclusion tests only on segments that can clear 25,000 matched after expected match rate. Use OpenAI’s 100,000 recommendation as the planning floor for serious inclusion tests, not as folklore.
- Separate the June CAPI path from the September audience path. If you already send conversions through LiveRamp’s CAPI Hub, document that as measurement. Do not let the new Custom Audiences ticket silently inherit conversion credit language you have not validated.
- Rewrite the client narrative. Approved line: LiveRamp can route RampID-matched first-party data toward ChatGPT Custom Audiences in a limited, unnamed market set, subject to OpenAI’s identifier and floor rules. Banned line: we have HubSpot-style CRM audiences everywhere ChatGPT sells ads.
- Assign owners. Identity and privacy own the key-crossing answer. Media ops own floors and bid adjustments. Analytics owns what “measure and optimize” is allowed to mean before LiveRamp publishes results. Creative does not own this ticket.
LiveRamp made ChatGPT Ads easier to plug into an identity stack many trading desks already run. That is real operational value. It is not proof that RampID is a first-class OpenAI audience key, not proof your country is covered, and not permission to brief CRM audiences on ChatGPT as if the match were already settled. The partner is not the match. Audit the key, the geos, and the floors before anyone spends against the logo.
