How should a service business run ChatGPT ads for leads?

The short version

A service business selling through enquiries, not a shopping cart, should treat ChatGPT ads the same way it treats Google and Meta: pick a lead event that represents a real intent to buy, feed the eventual sale back into the platform, and judge the result in a CRM rather than on the platform's own dashboard. The mechanics differ from search in the objective choice and the auction, but the discipline, capture the click, capture the lead, capture the outcome, is the same discipline that already runs your other channels.

Ad clickOpenAI adds a click reference called oppref to the landing page URL

Pixel in the browserFires page and conversion events and carries the click reference

Conversions API from your serverSends the same event with the same event ID, including sales that close later by phone or WhatsApp

Ads ManagerMatches the two by event ID so one sale is counted once, then reports and optimises on it

Choosing your objective

For most service businesses in a first test, oCPC is the more legible starting point, since the causal chain from click to lead to sale is easier to reason about than oCPM's broader exposure model. Move to oCPM later if your sales cycle genuinely involves people returning after seeing, but not clicking, an earlier ad. See our bidding strategies guide for the mechanics of both.

Choosing the lead event

OpenAI groups conversion activity into three stages, and its own guidance is that a healthy setup usually has events in more than one.

OpenAI is explicit that a completed outcome happening too rarely can be too sparse a signal on its own in the early weeks. Our approach for most service businesses: optimise the campaign toward the intermediate action, a lead submission, while feeding the completed outcome back separately as an offline conversion so you can still judge true return, even though the platform is optimising toward the earlier, higher volume step.

Feeding the real outcome back: offline conversions

This is the piece that decides whether a lead generation test on ChatGPT ads can actually be judged fairly, and it is the same pattern we use on every platform that bills on a click or a form fill rather than a closed sale.

1. Capture the click reference the moment a lead is created

   Store the oppref parameter, or whatever your CRM field maps to it, against the lead record itself, on whichever channel the lead arrives through.

2. Keep the original event timestamp, not the time you log the outcome

   A sale confirmed three weeks after the click needs the click's own timestamp preserved, not the date it was updated in your CRM.

3. Send the outcome through the Conversions API once it happens

   Tagged with the original click reference and event ID, so OpenAI can connect the eventual sale back to the ad that started it.

4. Test the full chain before trusting it

   From landing page, through the lead form or click to chat link, into the CRM, and back out to the Conversions API. Each handoff is a place the reference can be dropped.

The full mechanics of the click reference, advanced matching and the Pixel and Conversions API pairing are in our conversion tracking guide and ChatGPT ads pixel guide.

WhatsApp and phone handling

Most service businesses we see close on WhatsApp or a phone call, not on the landing page itself. If the only event the platform ever receives is a page view or a button tap, that is what it optimises to find more of, and a Conversions campaign will end up chasing button taps rather than sales. The fix has three parts specific to this pattern.

• Click to chat links need the click reference attached before the handoff. A WhatsApp click to chat button on your landing page is a moment the click reference can be captured, if your tracking is built to catch it before the user leaves your domain.
• Phone numbers need a way to trace back to the session. A dynamic number insertion setup or a click to call link that logs the click reference locally before dialling gives you something to match a later call against.
• WhatsApp Business and your CRM are often separate tools. If your WhatsApp workflow lives outside your CRM, someone has to move the click reference from the landing page, through the chat, and into the record that eventually triggers the Conversions API call. Map this handoff explicitly rather than assuming it happens automatically.

What your landing page needs

OpenAI's own guidance for landing pages applies directly here: link to the most relevant destination, not a generic homepage, and build a clear connection between the ad's promise and the page a lead lands on. For lead generation specifically, we would add two of our own points. First, the form or click to chat action should be visible without scrolling on a phone screen, since most ChatGPT usage on the unpaid tier and Go plan is mobile. Second, the page must not block OAI-AdsBot or OAI-SearchBot, since a landing page behind an aggressive firewall or a login wall will fail ad review entirely, not just underperform.

Writing ads and context hints for a lead generation offer

Context hints work differently to a keyword list here. Instead of naming your service category, describe the specific circumstance that makes someone a good fit right now. OpenAI's own framework is what, who and when: what you offer, who it helps, and when it is useful. For a service business, "when" tends to do the most work, since it captures the trigger moment a prospect is actually in when they start asking ChatGPT about a problem, rather than a generic category description. See our context hints guide for the full method and examples.

Build several ads per ad group, each taking a different angle on the same offer: one on price certainty, one on speed, one on a specific credential or guarantee. OpenAI's own creative guidance is that a high volume of distinct variations helps its system find more moments where your offer is genuinely relevant, and one ad per ad group is a habit worth leaving behind.

Sizing a budget for 15 or more leads

This is arithmetic on assumptions, not an OpenAI published benchmark, and your own click to lead rate is the number that matters, not ours. Use it to check a budget is realistic before you commit to it for four weeks.

If your own historical click to lead rate from Google or Meta is closer to 2 or 3 percent, the same budget produces fewer leads, and the test either needs more budget or more weeks to reach a judgable number. Put your own numbers into the budget calculator before committing, and see our cost guide for the full billing and budget pacing mechanics behind the AED figures above.

A four week test plan for lead generation

This is our own recommended plan, not an OpenAI benchmark.

1. Week 1: wire tracking and launch on Clicks

   Install the Pixel and Conversions API, confirm the click reference survives to your lead form, and launch a Clicks campaign with two or three offers while you validate the setup.

2. Week 2: confirm the lead event is firing correctly

   Check real leads are appearing as conversion events in Ads Manager, with the right event type and a click reference attached, before moving budget toward Conversions.

3. Week 3: move to Conversions once volume supports it

   Build a new campaign optimised toward the lead submission event, sized to gather enough volume to be meaningful rather than the platform minimum alone.

4. Week 4: reconcile against your CRM

   Compare cost per lead and, where sales have already closed, cost per sale, against the same period on Google and Meta, judged from your own CRM stages rather than each platform's own reported conversions.

Size the budget to reach at least 15 leads across the test. Fewer than that tells you very little about whether the channel works, whatever the daily minimum happens to be for your billing currency.

Common mistakes we see in lead generation setups

UAE and Gulf specifics

The pattern above matters more in the Gulf than in markets with a shorter, more digital sales cycle. Higher value services, business setup, property related work, private education, clinic consultations, can run days or weeks between a click and a closed sale, so preserving the original event timestamp through that gap is not optional. Build separate ads and context hints for Arabic and English rather than one bilingual creative, since OpenAI's ad unit carries a single title and copy per ad and language selection considers the user's location and settings alongside the conversation itself. If you operate in a regulated category, healthcare, financial services or legal services among them, confirm your lead capture and any claims in your ad copy also meet your sector regulator's requirements, since that sits outside OpenAI's own ads policy review. For the fuller account setup path before any of this runs, see ChatGPT ads in the UAE and how to advertise on ChatGPT.

Quick answers

Which campaign objective is best for lead generation on ChatGPT ads?

A Conversions campaign optimised toward a lead submission event once tracking is proven, started from a Clicks campaign if you are still setting up measurement. OpenAI requires at least one supported standard conversion event before a Conversions campaign can be created.

What event should a service business send as its conversion?

A lead submission, such as a form fill or a click to chat action, is the usual choice. OpenAI's own guidance is that a completed outcome that happens too rarely can be too sparse a signal on its own, so many advertisers also track an intermediate action alongside it.

How do I track a sale that closes on WhatsApp or by phone?

Capture the oppref click reference with the lead the moment it enters your CRM, then send the outcome back through the Conversions API once the lead becomes a real sale, tagged with the original click reference and timestamp.

Does a higher quality lead cost more on ChatGPT ads?

OpenAI does not publish a cost relationship between lead quality and price. What it does state is that conversion optimised campaigns need a real, supported event to optimise toward, and a poor quality or mismatched event can make delivery harder to judge.

Can I optimise toward a custom lead event, like a specific form on my site?

Not directly for a conversion optimised campaign. OpenAI states custom conversion events are not currently supported for optimisation; at least one supported standard event is required.

How long should a lead generation test run on ChatGPT ads?

Our view is a minimum of four weeks with a budget sized to reach roughly 15 or more leads, since a shorter or thinner test rarely produces enough volume to judge cost per lead fairly.

Should the landing page be the same for the ad and for Google or Meta traffic?

It can be, but a page built for the specific need in the ad group, rather than a general services page, is what OpenAI's own creative guidance and our own testing both point to for a stronger connection between ad and outcome.

Related