ChatGPT Ads: early results and what changes for customer acquisition

Low-cost clicks and a first demo below USD 10 sit alongside rapidly consumed budgets and poorly qualified leads. Early ChatGPT Ads results are mixed. For marketing teams, the practical question is where this channel belongs in their customer acquisition strategy.

We have examined several teams’ experiences to identify useful experiments, unresolved questions and the information teams need. Here is what those tests can teach us about reaching prospects and assessing commercial outcomes.

Platform capabilities checked on 8 September 2026. Figures come from separate tests, not market benchmarks. Costs retain their original currency: EUR for euros and USD for US dollars.

Key takeaways

ChatGPT Ads is OpenAI’s paid offering for companies that want to appear within ChatGPT. Sponsored placements are labelled and kept separate from the assistant’s answers. Paying for an ad does not change a response or buy an organic recommendation for your brand.

Someone using a search engine might type “walking boots”. In ChatGPT, they might explain their route, spending limit, season and concerns. Those details let the system select a relevant placement alongside the conversation.

Advertisers do not receive those private exchanges. Their job is to describe the situations in which their offer could be useful, then assess results using the information available in their analytics tools.

OpenAI began its US advertising pilot in February 2026 and subsequently expanded to additional markets, including the UK. Self-service Ads Manager access expanded across 31 further European markets on 31 August, including France. The eligible audience consists of people on Free and Go plans; higher-tier subscriptions remain ad-free.

That matters for B2B planning. The eligible audience is not the entire ChatGPT user base. How many colleagues use the assistant at work tells you little about how many of your potential buyers can actually see a campaign.

Campaigns can use text ads or formats built from a product catalogue. Buying and optimisation options include CPM, CPC and conversion-focused optimisation, subject to the choices available in the account. The interface brings together setup, budgets, targeting, creative and performance reporting in the account interface.

Keep three activities separate. SEO supports visibility in search engines; generative engine optimisation, or GEO, addresses visibility in AI-generated answers; ChatGPT Ads buys a paid placement. Uploading a sponsored product feed does not secure inclusion in organic answers.

Users also have control over personalisation. OpenAI keeps private conversations separate from the information shared with businesses. When planning a paid strategy on the platform, identify the situations in which your offer could be useful, alongside the settings and eligible formats available to you. Organic visibility and paid distribution can coexist, but neither guarantees the other.

Treat this platform as a distinct surface with its own experience for users. Familiar metrics can describe distinct buyer journeys.

A low CPC is not enough to validate the channel: track lead quality in your CRM.

Early reports include clicks around £0.30 to £1.50 across separate tests, but provide limited visibility into the opportunities and customers ultimately generated. Before increasing spend, check whether contacts fit your target customer profile and progress towards a sale.

Suggested bids are a starting point, not necessarily the price you need to pay.

Reported costs vary widely: some clicks came in around £0.30, while other tests recorded £4–5. An initial suggested bid of £5 was also reported. These figures come from different settings; start with a controlled bid and adjust according to delivery and lead quality.

An attractive cost per demo only becomes useful if it holds over time.

An experiment for one specific client spent £1,000 on awareness in two days without a conversion, then achieved a first demo at £9.29 in a clicks-led campaign before that cost increased. This highlights the importance of choosing the objective carefully; it establishes neither a stable average CPA nor customer acquisition cost. Track the cost of your chosen action across multiple conversions.

Describe the problem your offer solves, then assess the quality of the visits it attracts.

The available feedback on context descriptions suggests that action-oriented intent can achieve greater delivery than generic product categories. This is a hypothesis to investigate, not an established formula. Compare a few specific buying situations while keeping other settings constant, and assess qualified leads rather than impressions alone.

Give catalogue formats a dedicated test using available products and reliable information.

A two-month comparison for one client reported a £2.68 CPC and 1.8% CTR for the catalogue, against £3.99 and 0.9% for text. Conversions were reported only for the catalogue, without a count. That supports comparing formats on sales and CPA, rather than assuming a feed will always be more profitable.

Decide your acceptable cost per qualified lead or customer before launching.

Published results tell us more about CPC and click-through rates than final CPA or revenue. These reports therefore do not provide a sufficiently robust benchmark on which to base a profitability forecast. Compare the cost of completed actions, their conversion into sales and their value with your other channels.

The common lesson is to test a buying situation, objective or format, then connect expenditure to business outcomes. Observed costs inform your hypotheses; they are not a promise of success.

Context hints explain when an offer is relevant. They are not a guaranteed list of purchased search queries, and they do not reveal the prompts people actually used.

For instance, a reporting software company could describe “a marketing manager spending hours consolidating advertising results”. This gives the copywriting team a problem and a use case. It is a more specific starting point than “powerful marketing software”.

People exploring a problem are not necessarily ready to convert. An ad offering a comparison should lead to a resource that genuinely helps them compare. Advertising an available product should take the visitor to the correct item with consistent pricing.

A conversation provides a hypothesis about intent, not proof of commercial qualification. Check whether visitors match your customer profile and move towards purchase. If users repeatedly ask for services you do not provide, the problem may sit in the proposition rather than the bid.

Let's get back to our dashboard software: a demonstration of the workflow and time saved may be more useful than a generic website introduction. For walking equipment, the use case should remain clear all the way to checkout.

Comparing ChatGPT Ads with Google Ads, Meta Ads or LinkedIn Ads requires equivalent objectives, audiences and account maturity. A new discovery campaign should not be judged against established brand demand.

Look at new customers, qualified opportunities and customer acquisition cost. For a small experiment, distinguish advertising cost per customer from full CAC, which also includes other sales and marketing expenses.

Preserve the campaigns that already work while evaluating this new opportunity. It can earn a larger place in the acquisition plan once it produces sufficiently consistent commercial results, rather than simply a lower cost per click.

Organise measurement into three levels. Impressions, clicks, CTR, CPC and CPM describe delivery. Form submissions, purchases and qualified leads describe recorded outcomes. Incremental contribution refers to outcomes that would not have happened without the advertising.

Those levels are not interchangeable. A conversion attributed to ChatGPT may belong to a journey that also included an email, Google search and a sales conversation. Adding up the conversions claimed by multiple platforms can therefore overstate the total outcome.

Consistent marketing attribution rules make comparisons more useful. Align conversion definitions, time periods and qualification rules before drawing conclusions about channel performance.

In practice, retain the account’s view for delivery optimisation and a company-wide view for commercial decisions. They answer related questions, but will not always produce identical numbers. A genuine incrementality study needs an appropriate control and enough observations to support the conclusion.

A recorded interaction and an analytics session are separate events. An interrupted page load, redirects, consent choices, lost parameters or repeat clicks can all contribute to discrepancies.

Check destination URLs, UTM tagging, mobile loading and the information retained in the CRM. Then compare the same period and time zone. A persistent gap deserves investigation; it does not automatically demonstrate fraud or lost sales.

Useful diagnostic questions include whether the visitor reached the landing page, whether analytics recorded the visit and whether a later lead retained its acquisition source. Following the journey in that order helps avoid changing bids to compensate for a measurement issue elsewhere.

A pixel records events captured in the browser. The Conversions API lets an advertiser send events from its own systems. This becomes useful when a purchase or qualification happens later, outside the website or in the CRM.

Receiving an event does not automatically make it attributable. It must fit the measurement configuration and be connected to an eligible click within the relevant attribution window. Where two collection methods cover the same event, deduplication needs to prevent double counting.

Server-side measurement improves collection. It does not bypass consent, reconstruct private conversations or turn an unsuitable prospect into a valuable customer. Keep those limits clear when deciding whether the account needs better signals or a revised proposition.

DinMo supports all three ChatGPT Ads integrations: conversion delivery, product catalogue feeds and audience synchronisation. Each serves a different purpose: measuring commercial outcomes, showing relevant products and choosing which customers to include or exclude.

DinMo’s conversion connector sends online and offline conversion events from the data warehouse. Marketing teams select the relevant events and organise their synchronisation with the destination.

For B2B teams, an initial form submission may be followed by a qualified lead or an attended demo. For ecommerce, a completed order and its value tell you more than a product view. The appropriate choice depends on supported events, volume and how quickly the data becomes available.

Start with one reliable event, then extend coverage. Keep identifiers stable and reconcile sent volumes with the source records. Our guide covers how to send conversions to ChatGPT Ads with DinMo.

Give each event a precise operational definition. For example, a booked demo and an attended demo represent separate stages. Sending both under an ambiguous label makes it harder to understand the quality of the demand, even if the transfer itself works correctly.

Product formats depend on a structured feed. Prices, availability, descriptions, images and destination URLs should reflect what a customer can actually buy. An unclear description or an unavailable item can waste an otherwise relevant visit.

Choose a coherent product scope for initial experimentation: available items, sufficiently complete attributes and landing pages that work on mobile. The feed-versus-text observation above justifies comparing formats, not uploading every item indiscriminately.

DinMo supplies your advertising catalogue from product data in your data warehouse. Select the products to send, map their attributes to the required fields and schedule updates. This gives catalogue tests a reliable feed with current prices and availability.

The operational question is who owns data quality when an item changes. Agree responsibility for correcting prices, removing unavailable products and checking import failures. Automation is useful only when it keeps the customer-facing offer accurate.

ChatGPT Ads supports custom audiences for inclusion, exclusion and bid adjustments. Inclusion and bid adjustments currently require at least 25,000 matched users. Smaller audiences can be used for exclusions, a relevant limitation for specialist B2B customer bases.

An exclusion list can prevent recent customers from receiving an introductory offer. Value-based segmentation can also support decisions about which populations to prioritise, where audience sizes and available settings allow it.

DinMo builds segments from customer data and synchronises them with the advertising account. You can supply audiences and exclusion lists, keeping membership current as customers purchase or their profiles change. Check first-party data permissions and the platform’s audience size requirements.

For an acquisition initiative, define what counts as an existing customer before building the exclusion. For a reactivation initiative, use a separate eligibility rule. Clear definitions prevent teams from applying the same list to objectives that require distinct users to be reached.

Create an advertiser account, check eligibility and complete the required billing information. Select the objective, market, budget and bidding approach. Prepare the ad groups, descriptions of buying situations and landing pages. For a catalogue format, import the product feed and check the results before launch.

Validate measurement and customer journeys before submitting the creative for review. An agency can help with media strategy, ads and account settings. DinMo supports data activation. Establish those responsibilities early so that the owner knows where to turn when measurement or delivery needs attention.

Avoid launching a large collection of near-identical ad groups. With limited initial funding, splitting the available traffic too widely can leave every experiment short of observations. A smaller set of clearly distinct customer situations usually makes the learning easier to interpret.

The investment should allow useful observation without putting established channels at risk. As a planning example, aiming for 20 leads at an assumed £50 CPL implies £1,000 in expenditure. This is neither a universal budget recommendation nor a guarantee of statistical significance.

Choose one offer and a few use cases. Check early expenditure, then allow the normal conversion delay before assessing quality. In B2B, meaningful sales feedback may arrive after paid delivery has ended.

Your first objective is to establish whether this approach can attract valuable customers. DinMo supports the three data flows behind your tests: conversions to measure outcomes, catalogues to present products and audiences to include or exclude relevant customer groups. Speak to the team to set up these integrations.