Get Free Migration Kit

Salesforce Marketing Cloud Advertising Studio is no longer renewable. Run Cezium Ads alongside it. A simple install for better features!

SFMC Tips
8 min read

What Advertising on ChatGPT Actually Is — and What CRM Data Does There

Updated on
October 13, 2026
Categories and Tags
SFMC Tips
Industry News
SFMC
Marketing Cloud
Audience Matching
Cezium Ads Team
By subscribing you agree to our Privacy Policy.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
How it works — advertising on ChatGPT explained: formats, auction, 25k minimum, OAIQ pixel, oCPC

Yesterday we launched ChatGPT Ads as a Cezium destination. Today's post is the one we wished existed when we started building it: a plain, hype-free description of what advertising inside ChatGPT actually is — because most of what circulates is either OpenAI's careful consumer messaging or agency excitement, and a marketing team planning real budget needs the mechanics.

Everything below reflects the platform as of October 2026. It is early-days software; expect change, and check OpenAI's current documentation before committing spend.

Where the ads live

Ads appear inside ChatGPT conversations for users on the Free and Go tiers — clearly labeled, visually separated from the model's answers. Paid tiers (Pro, Business, Enterprise, Edu) see no ads at all. OpenAI matches placements to the conversation: the current topic, past chats, and past ad interactions — and deliberately keeps ads away from sensitive territory (health, mental health, politics and similar).

Two structural things follow from that design. First, context is the primary targeting signal — this is closer to search's intent model than to social's identity model, except the "query" is a conversation with real depth. Someone asking ChatGPT to compare CRM platforms is several layers more qualified than someone who typed "CRM" into a search box. Second, the audience skews toward the free consumer tier — which shapes which of your segments belong there (more below).

How buying works

The platform will feel familiar to anyone who has run modern paid media:

  • Three objectives: CPM (reach), CPC (clicks), and oCPC — conversion-optimized bidding where you're billed per click but the system bids toward a tracked conversion event, capped by a per-ad-group bid ceiling. One standard event per campaign; no custom events yet.
  • A relevance-weighted, second-price auction: your bid competes alongside the ad's contextual fit. Relevance isn't a quality-score garnish here; on a surface built entirely on conversation quality, it's load-bearing.
  • Structure: budgets and targeting at campaign level; bids and context hints at ad-group level. Self-serve Ads Manager in the launched markets, agency-partner access in most of Europe.
  • Measurement: the OAIQ pixel (OpenAI's measurement SDK) sets a 30-day first-party cookie on your site after ad clicks and reports page, commerce, lead, and subscription events; a Conversions API handles server-side events with deduplication. OpenAI generates the pixel per advertiser — no freeform tag soup.

The part that concerns us: custom audiences

ChatGPT Ads accepts advertiser-uploaded lists — emails or phone numbers, raw or SHA-256 hashed — matched privately against the user base (OpenAI never shows you who matched). Three uses: inclusion (target the list), exclusion (suppress it), and bid multipliers (spend more or less against it). Two constraints matter operationally:

  • A 25,000 matched-user minimum (OpenAI recommends 100,000+). This is a large-segment surface: customer bases, lapsed-buyer files, subscriber universes — not your 400-person VIP list.
  • Lists are immutable — no edits after creation. Every refresh is a new list. Manually, that's a chore that guarantees stale audiences; through an automated layer like Cezium, lists are recreated and swapped as the source Data Extensions change, which turns the platform's most awkward constraint into a non-event.

What CRM data is actually for on this surface

Here's our honest read, having built the pipe: on a context-driven surface, first-party audiences matter defensively before offensively.

  • Exclusion is the killer app, day one. The conversation targeting will find people discussing your category — including your existing customers asking support-flavored questions. Suppressing your customer base makes every paid impression net-new and keeps acquisition spend from harassing the already-acquired. This alone justifies connecting the pipe.
  • Inclusion works for the big lifecycle motions. Win-back across a lapsed file, upgrade messaging across a free-tier user base — segments that clear the 25k bar and gain from meeting people in a considered, conversational moment.
  • Bid multipliers are the subtle play. Bid up when the conversation context finds someone on your high-propensity list; bid down against recent purchasers you don't want to exclude entirely.

And one caution we'd rather publish than have you learn: this is not yet a precision-lifecycle channel. Minimum sizes, immutable lists, single-event optimization, young reporting — the platform is built for broad strokes today. Run your surgical segments on Meta and Google; bring your big, clean, well-consented files here; measure everything; scale on evidence. Being early is an advantage only when paired with the discipline to stay honest about results — that's the same CRM-team discipline that makes first-party audiences outperform everywhere else, applied to the newest surface there is.

Frequently asked questions

How do ads on ChatGPT work?

Clearly labeled sponsored placements appear in conversations for Free and Go-tier users, matched to conversation topic and history via a relevance-weighted second-price auction, with CPM, CPC, and oCPC buying objectives.

Who sees ads in ChatGPT?

Free and Go-tier users in OpenAI's launched markets. Pro, Business, Enterprise, and Education subscribers see none — and ads avoid sensitive topics like health and politics.

Can you target ChatGPT ads with your own customer lists?

Yes — hashed email/phone lists with at least 25,000 matched users, usable for targeting, suppression, or bid adjustment. OpenAI matches privately and never exposes individual users.

How do you measure ChatGPT ad performance?

Through the OAIQ pixel (30-day first-party cookie, standard event types) and a server-side Conversions API with deduplication, plus an Insights API for programmatic reporting.

What should a brand test first on ChatGPT Ads?

Customer suppression, then one large-segment inclusion play (win-back or upgrade), with a modest budget and conversion tracking in place from day one.

Put your first-party data to work on ChatGPT — talk to us →

Mounir Nejjai
About the author
Mounir Nejjai
Founder & CEO, Cezium

Mounir Nejjai is the founder and CEO of Cezium, the SFMC-native audience activation platform. A recognized Salesforce Marketing Champion, he helps enterprise marketing teams move first-party data out of Salesforce Marketing Cloud and into the ad platforms where it drives revenue. He writes on audience activation, SFMC advertising, and the post–Advertising Studio landscape. Based in Paris.

Connect on LinkedIn →

Ready to transform your CRM audience activation?

Join marketers who've simplified their workflow with Cezium Ads