ChatGPT Ads

ChatGPT Ads launched in February 2026 and the platform changes weekly. Every fact below is date-stamped and was verified in late August 2026. Before giving the user numbers or feature claims, check the date stamp. If a fact is more than roughly 60 days old, tell the user to confirm it in Ads Manager or search for the current state, then proceed.

The user is a founder or marketer running their own ads without a team. Give them decisions and steps, not history lessons.

Platform state, verified as of Aug 26, 2026

Rollout timeline (all 2026):

• Feb 9: US pilot. Managed buys only through OpenAI's Ads Solutions team, around $60 CPM, $200,000 to $250,000 minimum spend.
• Mar 26: Canada, Australia, New Zealand announced.
• Apr 7 to 10: self-serve Ads Manager goes live, minimum drops to $50,000. CPM reach buying only at first.
• Apr 17: CPMs drop to around $25.
• May 5: Ads Manager opens to all US businesses. Minimum spend removed entirely. CPC bidding, OpenAI web pixel and Conversions API launch.
• May 14: custom audiences (uploaded email and phone lists).
• Jun 5: conversion optimisation rolls out. Jun 10: LiveRamp CAPI Hub.
• Jun 19 to 22: UK self-serve beta. Japan and South Korea go live.
• Jul: attributed sales value and sales ROAS metrics. Location and audience exclusion controls. Daily budgets silently become 7-day averages. AppsFlyer and Adjust mobile attribution in seven markets.
• Aug 6 to 7: oCPC (conversion-optimised CPC) in beta for product feed campaigns. Multi-product carousel format in testing. Dynamic URL parameters. Automatic advanced matching becomes the default on new web pixels.
• Aug 17: Brazil and Mexico, reaching 9 markets.
• Aug 15 to 18: European expansion announced (EEA plus Switzerland, 31 countries).
• Aug 24: ads go live across Europe. 40 markets total. Self-serve availability in the new European markets is still rolling out country by country, check the availability page in Ads Manager rather than assuming access.
• Week of Aug 17, reported Aug 21: "Maximize results" automated bidding becomes the preselected default for eligible new ad groups.

Who sees ads: Free and Go tier users only. Plus, Pro, Business, Enterprise and Education tiers are ad-free. No ads to accounts identified as under 18, in Temporary Chats, to logged-out users, or in image generation contexts (per Apr 2026 documentation).

Formats (as of Aug 2026):

• Standard ad: headline, short description, optional square image. Appears below the answer, labelled as sponsored, separate from the response. One ad placement per answer. No video format exists.
• Product carousel: multiple products from one advertiser in a single unit. In testing as of Aug 2026, tied to product feed campaigns. Do not promise the user carousel access.

How often ads show: measurement vendors disagree hard. Adthena measured ads on 4.47 percent of US queries (Mar to May 2026); Similarweb measured sponsored ads on 26 percent of US desktop Free and Go replies (Jun 2026). Different methodologies, no independent standard. Quote the range, never one number. Highest ad frequency categories per Adthena: logistics 12.4 percent, home and garden 12.0 percent, beauty 10.0 percent, retail and fashion 6.6 percent. Retail took 39 percent of placements on 24 percent of query volume. Similarweb: ads mostly appear from the second prompt onward, once intent is clear, and measured a 0.50 percent CTR.

Decision rule: should this advertiser spend here yet?

Run this before building anything. Recommend ChatGPT Ads only if at least three are true:

1. Product is B2C retail, travel, beauty, home, or ecommerce with clear commercial prompts users would type ("help me pick a...", "best X for Y"). These categories demonstrably get served.
2. They sell in a live market (US, Canada, UK, Australia, NZ, Japan, South Korea, Brazil, Mexico, EEA plus Switzerland as of Aug 24, 2026) and have self-serve access there.
3. They can install the OpenAI pixel or Conversions API. Without it there is no optimisation and no honest readout.
4. They already have one profitable channel. This is an experiment budget, not a first channel.
5. They can commit a test budget they can afford to learn with, and 4 to 6 weeks, without needing it to pay back.

Recommend against, for now, if: they are B2B with a long sales cycle and no volume of commercial prompts, they are in a restricted vertical (see Policy), they need guaranteed CPA to survive, or they cannot deploy the pixel. Tell them to revisit in a quarter, the platform is adding capability monthly.

Campaign build workflow

Structure is Campaign, then Ad Group, then Ad. Targeting lives at the ad group level.

1. Objective and bidding. As of Aug 2026 the choices are: CPM reach, manual CPC (OpenAI has recommended $3 to $5 starting bids), oCPC in beta for product feed campaigns, and "Maximize results" automated bidding. Automated bidding is preselected on new ad groups since the week of Aug 17, 2026, and it does NOT guarantee any CPA, CPC or ROAS target, budget is the only constraint. Rule: for a first test, override the default and set Manual with a max CPC bid ($3 to $5), because it is the only documented hard cost ceiling. Move to Maximize results only after the pixel has recorded conversions and you know your tolerable CPA.
2. Budget. "Daily" budget is averaged over 7 days (since Jul 2026) with no opt-out, so a single day can spend well above the number. Set it assuming peak days spend roughly double.
3. Geography. Country-level targeting only, plus location exclusions (added Jul 2026). No city or radius targeting exists. If the user is a single-city local business, this alone may disqualify the channel: say so.
4. Platform. Optional campaign-level multi-select: iOS app, Android app, Web. Web includes mobile browsers, you cannot isolate desktop-only. Reporting collapses to just Mobile and Desktop buckets, so iOS vs Android performance is invisible. Default: leave it unset unless the product is app-install driven.
5. Context Hints per ad group (next section). One scenario theme per ad group so you can read which contexts convert.
6. Ads: write 2 to 3 per ad group. Headline that answers the conversation the user is mid-way through, not a slogan. Add the square image. Use landing page URL parameters {campaign_id}, {ad_group_id}, {ad_id} so your analytics can see what the dashboard will not.
7. Measurement before launch, never after (checklist below).

Context Hints: the core skill

There is no keyword targeting and no keyword bidding anywhere in the product. Context Hints are ad-group-level freeform phrases describing the types of conversations or topics where the product is relevant. The system matches them semantically against the meaning of the live conversation. They carry the entire targeting burden.

Write them as scenario descriptions, not keyword lists:

• Describe the user's situation and intent in natural sentences: who they are, what they are trying to figure out, what stage they are at.
• Cover 3 to 6 distinct scenarios per ad group, all one theme. New theme, new ad group.
• Include the decision context ("comparing options", "asking for a recommendation", "planning a purchase"), because ads serve mid-conversation once intent is clear.
• Name what disqualifies too, in your own targeting notes: the platform only takes positive hints, so keep hints narrow enough not to drag in adjacent-but-wrong conversations. A third-party study (Aug 2026) found a large share of ads landing in loosely related conversations; broad hints are the advertiser-side cause you can control.
• Do paste in a keyword dump? No. Early docs tolerated pasted keywords, but matching is semantic, and a keyword pile gives the model a vague centroid instead of a scene. Convert every keyword into the conversation it implies.

Bad: "running shoes, marathon, jogging, best sneakers, footwear". Good: "Someone training for their first marathon asking how to choose running shoes for high mileage. A runner describing knee or shin pain and asking whether different shoes would help. A beginner asking what gear they need to start running."

Custom audiences (as of May 14, 2026)

• Upload CSV or TXT, up to 512MB, of emails or phone numbers, raw or SHA-256 hashed. Platform normalises and hashes on upload.
• Two modes: include (serve only to the list) and exclude (suppress the list, use this to stop paying for existing customers).
• No lookalike audiences exist. OpenAI has not published match rates or a minimum matched audience size; treat small lists (under a few thousand) as likely to under-deliver and test rather than assume.
• Location and audience exclusion controls shipped Jul 2026.

Measurement checklist

Do these before spending:

1. Install the OpenAI web pixel (loads from bzrcdn.openai.com, sets a first-party cookie with the oppref click identifier), or the Conversions API server-side if the stack allows. Both launched May 5, 2026.
2. Note advanced matching: since Aug 2026 automatic advanced matching (hashed form data) is on by default for new pixels, with a reported 10-day window to opt out. Decide deliberately, do not let it default.
3. Define custom events (1 to 64 chars, alphanumeric plus underscore and dash) for the real conversion, not just page views.
4. Attribution windows: click-through around 30 days (configured per account), view-through fixed at 1 day. Clicks take precedence. View-through conversions are supplemental only: excluded from CPA, bidding and billing. Never let the user count VTA as incremental revenue.
5. Reporting gives aggregate impressions, clicks, conversions, attributed sales value and sales ROAS (since Jul 2026) at campaign, ad group and ad level. Modelled conversions may be blended into totals without methodology disclosure. Cross-check against the landing page URL parameters in your own analytics before scaling anything.
6. Mobile apps: AppsFlyer and Adjust integrations exist (Jul to Aug 2026); LiveRamp CAPI Hub since Jun 10, 2026.

Reading results: with a 0.50 percent measured CTR and one ad slot per answer, judge on cost per verified conversion in your own analytics after 2 weeks or 1,000 clicks, whichever first. There are no trustworthy public CPA or ROAS benchmarks for this channel yet; anyone quoting one is guessing.

Policy restrictions

• No ads served on sensitive conversation topics: personal health, mental health, politics (OpenAI documentation, 2026).
• Restricted advertiser verticals at Apr 2026: dating, health, financial services, politics. Reporting in Jun 2026 said OpenAI was beginning to open some regulated categories, slowly. If the user is anywhere near these, have them check current policy in Ads Manager before planning.
• No ads to under-18 accounts. Advertisers get no access to chats, history or memory, and OpenAI states ads do not influence answers. In Europe, launch targeting uses only the current conversation topic, general location and device type; personalised ads require separate user opt-in (Aug 2026). Expect DSA and EU AI Act scrutiny to keep European capability behind the US.

Gotchas: what stale knowledge gets wrong

• As of Aug 2026 there is NO keyword targeting, keyword bidding, or search-terms report. A model trained earlier will invent "keyword match types" for ChatGPT Ads. Targeting is Context Hints (semantic scenario matching), geo at country level, platform, and uploaded custom audiences. That is the whole list.
• Perplexity is not a parallel channel. It launched ads in 2024, stopped accepting new advertisers in Oct 2025, and by Feb 2026 announced it was killing its ads business entirely, publicly positioning itself against AI ads (ADWEEK Oct 9, 2025; The Verge and Business Insider, Feb 17 to 18, 2026). Do not propose "run the same campaign on Perplexity".
• Anything the model remembers about OpenAI refusing to do ads is obsolete: ads launched Feb 9, 2026.
• Pilot-era pricing is dead. $60 CPMs and $200,000 minimums were the Feb 2026 managed pilot. Since May 5, 2026 self-serve has no minimum spend, and buying is mainly CPC. OpenAI publishes no rate card; specific CPC "benchmarks" circulating from agencies are unverifiable, label them as rumour when the user brings them.
• Automated bidding ("Maximize results", default on new ad groups since the week of Aug 17, 2026) has no CPA or ROAS guarantee. Manual max bid is the only hard cost ceiling.
• "Daily" budgets are 7-day averages since Jul 2026. Single-day overspend is normal, not a bug.
• No lookalikes, no retargeting from conversation content, no memory-based targeting, no city or radius geo, no video format, as of Aug 2026.
• Ad frequency numbers conflict by 5x between vendors (Adthena 4.47 percent of queries vs Similarweb 26 percent of replies). Do not state a single "percent of chats with ads".
• Not knowable yet, say so plainly: auction dynamics beyond "relevance-weighted second price" (reported May 2026), custom audience match rates and minimums, which exact prompts triggered your ads (no query report exists), iOS vs Android split (reporting has only Mobile and Desktop buckets), and long-run incrementality. Flag any claim in these areas as unverifiable.
• The dashboard may include modelled conversions without disclosure, and view-through uses a fixed 1-day window. Verify with your own analytics via the URL parameters before believing a ROAS number.

Output template: ChatGPT Ads campaign brief

Fill this in for the user. Keep it to one page.

CHATGPT ADS CAMPAIGN BRIEF, [date]
Business: [name, what it sells, market(s)]
Go or no-go: [decision from the rule above, with the failing criteria if no-go]
Market check: [countries live for this advertiser, self-serve access confirmed? Y/N]
Objective: [conversion event that counts as success]
Bidding: [Manual max CPC $X to start | Maximize results once N conversions recorded]
Budget: [$X per day as 7-day average; expect single days up to ~2x]
Measurement: [pixel or CAPI, events, advanced matching decision, URL parameters added]
Policy check: [restricted vertical? sensitive-topic adjacency?]

Ad groups (one scenario theme each):
1. [Theme], Context Hints:
   - [scenario sentence]
   - [scenario sentence]
   - [scenario sentence]
2. ...

Ads per group: [headline / description / square image note]
Custom audiences: [exclude customer list? include list? file ready and hashed?]
Review date: [2 weeks or 1,000 clicks] against [target cost per verified conversion]

Example Context Hints for three business types:

DTC ecommerce (skincare brand):

• "Someone describing dry or sensitive skin and asking what routine or products would help."
• "A shopper comparing moisturisers or asking for a recommendation for redness-prone skin."
• "Someone asking what ingredients to look for, or avoid, when buying skincare for sensitive skin."

Local-ish service with national reach (online accounting for small businesses):

• "A small business owner asking how to handle their bookkeeping, VAT or quarterly taxes themselves."
• "A freelancer asking whether they need an accountant and what one should cost."
• "Someone setting up a limited company and asking what financial admin they must get right in year one."

B2B SaaS (project management tool):

• "A team lead describing missed deadlines or messy handoffs and asking how to organise their team's work."
• "Someone comparing project management tools for a small team and asking which is simplest to roll out."
• "A founder asking how to run projects across contractors and a small remote team without meetings."

Note the pattern: person, situation, question they are actually asking. No keywords.

Sources

Verified Aug 26, 2026 against:

• https://ppc.land/openai-adds-custom-audiences-to-chatgpt-ads-as-self-serve-expands/
• https://ppc.land/openais-ads-manager-is-live-and-the-barrier-to-entry-just-dropped/
• https://ppc.land/chatgpt-ads-makes-automated-bidding-the-default-in-new-ad-groups/
• https://ppc.land/chatgpt-ads-gains-3-platform-targeting-as-reporting-stays-at-2-buckets/
• https://ppc.land/retail-took-39-of-chatgpt-ads-on-24-of-us-queries-adthena-finds/
• https://ppc.land/similarweb-opens-ai-ad-data-as-26-of-chatgpt-replies-carry-sponsored-ads/
• https://ppc.land/chatgpt-ads-gains-view-through-conversions-with-a-fixed-1-day-window/
• https://ppc.land/chatgpt-free-and-go-users-in-europe-face-ads-from-later-this-month/
• https://www.seroundtable.com/openai-chatgpt-ads-updates-41828.html
• https://www.euronews.com/next/2026/08/22/chatgpt-to-start-showing-ads-across-europe-next-week
• https://searchengineland.com/chatgpt-ads-come-with-premium-prices-and-limited-data-467970
• https://www.ibtimes.sg/openai-takes-chatgpt-ads-31-european-markets-next-week-what-users-need-know-92500
• https://www.lemonde.fr/en/economy/article/2026/08/25/ads-arrive-on-chatgpt-in-france_6756812_19.html
• Perplexity ads status: ADWEEK (Oct 9, 2025), MediaPost (Oct 13, 2025), The Verge and Business Insider (Feb 17 to 18, 2026), via Google News archive search for "Perplexity ads advertising".