What is the minimum effective budget for the first 90 days?

Whatever buys you daily variance you can read for at least four weeks. The exact dollar amount is category-dependent; for most DTC brands today it is meaningfully smaller than monthly Meta spend. The error is sizing the budget as a percent of an existing channel rather than as a function of learning velocity. Size it for learning.

How many products or categories should I launch with?

Two to four. Fewer than two and you cannot diff the model's behavior across categories; more than four and you spread learning attention thin during the period when attention is the constraint. The brands that try to launch ten categories at once typically learn slower than the brands that launch three and iterate.

What KPIs should I report on internally during the first quarter?

Total-portfolio metrics, not channel-isolated ones. Blended CAC, repeat-purchase rate, gross profit per visitor, revenue mix across surfaces over rolling 30-day windows. Channel-level ROAS on OpenAI Ads will read low under most attribution setups; do not report it to the board as if it were the whole story.

What happens after day 90 if the channel did not produce clear winners?

Almost certainly the issue is brand-authority maturity in your category, not channel viability. Pause the ad spend, continue the brand-authority work for another quarter, and restart with the same playbook on day 180. We see this pattern often, and the brands that hold the playbook through that gap end up surfacing later than they expected, but on a stronger long-term footing.

Should I do this in-house or hire an agency for the first 90 days?

In-house, if you can. Agencies are good at executing inside known auctions; new surfaces reward in-house learning loops because the operator who reads the model's recommendations every week builds an intuition that an agency cannot transfer back to you cleanly. The exception is brands without the capacity to run any new channel themselves; in that case, an agency relationship structured around a 90-day learning sprint, not an open-ended retainer, makes sense.

How does this playbook integrate with my existing Meta and Google work?

It runs in the same planning loop, not as a parallel track. The brands that compound advantages across surfaces run all channels through one weekly review, one creative-iteration cadence, and one set of portfolio metrics. The siloed alternative is a step backward that the comparison piece OpenAI Ads vs Meta Ads makes in more detail.

The First 90 Days of OpenAI Ads: A DTC Playbook

The 90-day arc, in one frame

Three phases, three jobs. Each phase prepares the next; skipping one costs you the rest.

Days 1–30

Get on the channel

Be present with creative the model can use.

Days 31–60

Find the learning loop

Identify which combinations the model surfaces, and why.

Days 61–90

Scale what is working

Double down on winners, retire what did not produce learning.

The first 90 days on any new ad channel are worth more than the next 270 combined, because that is the period when the auction is undersold and your team's learning is uncompounded by anyone else's. Run those 90 days well, and your unit economics on the channel stay better than your competitors' for two or three years. Run them badly, or wait, and you spend the next two or three years catching up to brands who used the window correctly.

This post is the operating playbook. It assumes you have already decided to be on OpenAI Ads, which is the argument we made in the pillar explainer and the platform-shift thesis. What follows is concrete: what to do in days 1 through 30, days 31 through 60, days 61 through 90, what to avoid, and how to measure when the attribution data is incomplete by design. The advice is opinionated. Some of it will not match your category exactly; the structure should.

Days 1 to 30: get on the channel

The single goal of the first 30 days is to be present on the channel with creative that the model can use. Not to scale. Not to optimize. Not to attribute. Be on, with creative that survives paraphrase, in your core categories. Everything else compounds from there.

Week one is connection and infrastructure. Three concrete checks before anything else. First, audit your robots.txt and confirm `OAI-SearchBot` is allowed; blocking it makes your products invisible to ChatGPT's recommendations regardless of ad spend. Second, install OpenAI's Conversions API and pixel, which launched in May 2026 and is now the measurement baseline for ChatGPT-originated buyers landing on your site. Third, audit product data: JSON-LD Product schema, server-side rendered, with complete attributes, accurate pricing, current availability. The model will surface your product based on what your catalog says about it; if the catalog is thin, the model has nothing to work with. Spend the week making your product data legible.

Week two is creative prep. Write three core positioning statements for each product category you intend to advertise. Each statement should answer three questions: what the product is, who it is for, and why it is the right choice over the next-best option. Specificity beats emotion in this format. The brand that says 'vitamin C serum, 15% concentration, formulated for sensitive skin, third-party reviewed across 12,000 verified buyers' gets surfaced; the brand that says 'glow, radiance, transformation' does not.

Week three is launch with a learning budget. Pick two to four product categories, fund each with enough impressions to teach you something, and turn the channel on. Launch with cost-per-click bidding in the three-to-five-dollar range per Digiday's reporting on the April 2026 CPC rollout. The minimum effective budget is whatever buys you enough conversational impressions to see daily variance, not a percentage of any other channel. Most DTC brands underspend at launch because the dollar amount looks small relative to Meta; that is the wrong comparison. The right comparison is whether the spend produces a daily signal you can read.

Week four is hypothesis-setting. Write down, in a shared doc, what you think will work in the next 60 days and why. Be specific. 'I think the Replacement Shopper archetype will outperform the Researcher in our category because our repeat-purchase rate is high' is a useful hypothesis. 'I think this channel will work' is not. The hypothesis doc becomes the artifact you measure against in days 31 to 60.

Days 31 to 60: find the learning loop

The second 30 days are about building the feedback loop that will compound for the next 12 months. The goal is not to scale spend. The goal is to identify the two or three creative-and-category combinations that the model is consistently surfacing, and to understand why.

Audit which products are getting recommended. Once a week, sit with the team for an hour and ask ChatGPT the queries you think your buyers ask. Note which products surface, in what order, with what reasoning. This is qualitative work, and most analytics tools will not give you this signal. The hour you spend doing it manually is the most valuable hour of the week during this period.

Compare against your hypothesis doc. Where you predicted correctly, double down. Where the surface is recommending you in a way you did not expect, lean into it. Where you expected to be surfaced and are not, the gap is almost always one of two things: your product data is thin in a way the model needs, or your category authority is weaker than you assumed. Both are fixable; both take more than two weeks. Start fixing now.

Tighten the creative. The first 30 days produced creative that worked in the format; the next 30 days are about which specific positioning statements are surviving paraphrase well and which are not. The model's recommendations are themselves data. If the model is rephrasing your product positioning in a way that loses your differentiation, your positioning is not yet specific enough. Rewrite. Test. Repeat.

Build the brand-authority signals that compound. The work that makes you more recommendable on OpenAI Ads is the work that makes you more recommendable across every AI surface, not just ChatGPT. Reviews, structured data, expert endorsements, third-party validation. We listed the seven concrete moves in the agent-proof brand piece; most of them take 30 to 90 days to compound into model behavior. Start the long-lead ones now.

Re-baseline measurement. By day 45, you should have enough data to see the gap between channel-reported performance and actual revenue attribution. If your last-click analytics is reading the channel as 'underperforming,' the analytics is wrong. Most channels read low under last-click attribution; this one reads especially low because a meaningful share of conversions cross channels on the way to checkout. Switch to a multi-touch view, or at minimum read total-portfolio metrics rather than channel-isolated ones.

Days 61 to 90: scale what is working

The final 30 days of the first quarter are when you scale the categories and creatives that worked, retire the ones that did not, and expand the channel into the surfaces that compounded faster than expected.

Increase spend on the winning categories. The right increase is enough to capture more inventory without flattening the variance you are reading from the model. A doubling of spend is often the right move at this stage; tripling is sometimes too much, because the auction in your category may not have the inventory at that level yet. Watch for the point where additional spend stops increasing recommendation volume; that is the current inventory ceiling, and pushing past it wastes budget.

Retire what did not work. Not every category will succeed in 60 days; some will need a longer brand-authority buildout before the model starts surfacing you. The right call is to pause the spend, not to keep funding a category that has not produced learning. The brand-authority work continues regardless of the ad spend, and the ad budget refocuses on the categories where the model is already cooperating.

Expand creative range. The first 60 days produced two or three winners. The next 30 days are about understanding why those winners worked so you can produce variants. Is it the specificity? The trust signals? The product configuration? The pricing positioning? Test variants that change one variable at a time, not creative overhauls. The format rewards careful iteration, not creative pivots.

Set the next quarter's hypothesis. The 90-day playbook produces an artifact, which is a clearer hypothesis for the next quarter than you had at week one. By now you know which archetypes (Researcher, Replacement Shopper, Discovery Shopper) are predominant in your categories. You know which creative patterns the model favors. You know where your brand-authority gaps are. The next quarter's plan is built on those answers, not on industry templates.

What to not do

Seven anti-patterns that show up in almost every brand's first 90 days on a new channel.

Anti-patterns to avoid

Do not chase channel-level ROAS in the first 60 days. Attribution is incomplete, the auction is unsettled, and the ROAS number will be wrong in both directions for weeks. Track total-portfolio metrics, not channel-isolated metrics, during this window.
Do not run a $5K test for two weeks and conclude the channel does not work. That is not a test, that is a sample-size error. The minimum useful learning budget is whatever buys daily variance for at least four weeks. Below that you are reading noise.
Do not treat OpenAI Ads as a Meta lookalike. The auction is different, the creative discipline is different, the buyer journey is different. Importing the Meta playbook wholesale will produce confused results and waste creative budget on patterns the format does not reward.
Do not silo the channel. A separate dashboard, a separate team, a separate Friday spreadsheet, all of these slow the learning loop and prevent compounding with your other surfaces. Run it inside the same planning loop as Meta and Google.
Do not over-rotate after one good week. New channels produce variance. A week where everything seems to work is sometimes the auction temporarily favoring you for category-specific reasons that do not persist. Hold the playbook for 30 days before declaring a winner.
Do not over-rotate after one bad week. Same logic in reverse. The channel will have weeks where the model's recommendations look thin. The instinct is to cut spend or change creative; the right move is usually to wait and see whether the next week's data confirms the pattern.
Do not skip the brand-authority work. The ad budget is the visible spend, but the long-compounding asset is the work that makes you more recommendable across every AI surface. Skipping that work to focus on creative iteration is optimizing the wrong variable. Run both in parallel.

Measurement during the dark window

The hardest part of the first 90 days is measuring honestly while the attribution data is still maturing. The May 2026 launch of OpenAI's Conversions API and pixel materially closed the gap from the prior survey-only world, but no single instrument is enough on its own. Three measurement principles get you through it.

Triangulating channel signal in the dark window

No single instrument is reliable on its own. The honest read sits at the intersection of three imperfect signals.

Signal 1

OpenAI Conversions API

Pixel fires on the merchant-side conversion page. The May 2026 launch put native attribution inside reach for the first time.

Signal 2

Landing-page UTMs

Properly UTM-tagged landing pages catch the click-throughs that the pixel might miss, and give your analytics a coherent attribution path.

Signal 3

Post-purchase survey

One question, one minute. The long-tail safety net for the buyers neither pixel nor UTM caught.

Decision read

Trust the intersection. Do not trust any single signal alone.

Switch to total-portfolio metrics. Blended CAC, repeat-purchase rate, gross profit per visitor, and revenue mix across surfaces. These metrics absorb the noise of single-channel attribution and let you read whether the addition of OpenAI Ads is helping the whole, regardless of what any single channel's last-click number says. If your blended CAC drops while OpenAI Ads is on, the channel is doing its job, even if the channel-isolated ROAS reads as flat.

Simulate before you scale. Before significant spend increases, run scenario simulations to understand the expected lift band. We covered the discipline in know before you spend, and the principle is the same here: do not push budget toward a hypothesis you have not stress-tested first. The cost of stress-testing is minutes; the cost of overspending against a wrong hypothesis is the rest of the quarter.

Use post-purchase signal. The cheapest measurement instrument during the dark window is a one-question post-purchase survey: 'How did you first hear about us?' with ChatGPT, Search, Social, Friend, and Other as options. The signal is imperfect, but it is signal. After 100 purchases you have a directional read on the channel mix that no analytics tool will give you cleanly for another year. Build the survey in week two; you will be glad you did by week eight.

Resist the urge to optimize on noise. The first 30 days of any new channel produce wild swings. Variance is the signal during this period, not the signal to react to. Set your decision rules in week one and stick to them, because the decision-quality penalty of mid-stream reactive adjustment is higher than the cost of any specific week's underperformance.

Ninety days is enough time to learn a new channel well enough to compound on it for the next two or three years. Run the playbook. Build the brand-authority asset that pays off across every AI surface. Resist the over-reactions in both directions, and trust the portfolio metrics over the channel-isolated ones. If you want to start now, the OpenAI Ads page shows the integration path, and you can try Cresva for free to fold the new channel into the same planning loop your team already runs.

First 90 days, executed from one planning loop. Cresva agents handle the connection, the planning, the iteration, and the reporting. You run the playbook, the agents handle the operations.