Advertising on ChatGPT should start with a business question: can this channel produce customers you can profitably serve? Not whether a campaign can generate clicks. Not whether a screenshot looks impressive.
For a US local service business, that means inquiries inside the actual service area. For B2B software, it means buyers whose needs, company size, and buying timeline fit the product. Both need a traceable route from ad to landing page to accepted lead to sales outcome.
This is a US campaign and lead-tracking guide, based on official documentation checked on October 8, 2026. The examples are planning illustrations, not results from campaigns we have run. Account access, eligibility, and available settings must be checked in your own account before committing a budget.
Start with a controlled test brief
Before opening an ad builder, write a brief small enough that someone else can explain the test without a meeting.
For an illustrative HVAC company serving part of the Phoenix metro area, the question might be: “Can replacement-estimate inquiries from ChatGPT turn into qualified appointments within our service territory?” This is different from buying traffic for every service the company offers.
For an illustrative workforce-scheduling software company, the question might be: “Can operations managers at multi-location service businesses request relevant demos?” A student researching scheduling tools and a buyer responsible for 15 locations should not look identical in the results.
Record these decisions:
- Buyer: who should respond, and who would be a poor fit?
- Problem: the decision the buyer is trying to make.
- Offer: an estimate, demo, assessment, or other concrete next step.
- Geography: where the business can actually sell and deliver.
- Primary event: the accepted action you intend to measure.
- Quality rule: what makes that response commercially useful.
- Spending boundary: an approved exposure limit and named owner.
- Review point: when you will judge the test, including sales-cycle delay.
- Stop condition: what requires a pause even if clicks are arriving.
Keep one offer and a limited set of creative variations. If you change the audience, offer, landing page, and bidding together, you lose the ability to explain the result. A controlled test is useful even when it fails: it tells you which assumption did not hold.
Our ChatGPT Ads service starts with this connection between campaign design and business measurement. The campaign is only one part of the work.
Understand the account, campaign, ad group, and ad
OpenAI documents a four-level structure: the account contains campaigns, campaigns contain ad groups, and ad groups contain ads. Campaigns own decisions such as objective and budget; ad groups hold bidding and context hints; ads provide the creative and destination. Official API overview.
Use that structure to keep the test legible. Give a local estimate campaign and a national software-demo campaign separate budgets and reporting. Group related creative variants together when their bidding and context are shared.
Use names that help humans and retain IDs for reliable reporting. A name such as “US / Scheduling / Demo / October Test” is easier to interpret than “Campaign 7.” Keep a short change log containing the date, old setting, new setting, and reason. Do not rename losing ads to hide their history.
Prepare resources paused. Preview the ad, inspect account and creative reviews, and check the final destination before enabling delivery. A preview confirms appearance, not that every serving requirement is satisfied. Official campaign management guide.
Choose the objective without confusing it with billing
Current documentation supports impression, click, and conversion objectives. Impression campaigns use impression billing and fixed bids. Click and conversion campaigns use click billing. Fixed bidding supports daily or lifetime budgets; maximize-clicks and maximize-conversions strategies use daily budgets for their corresponding objectives. Official bidding and budgets guide.
The business implication matters: a conversion objective does not mean you pay only when a qualified customer appears. A conversion campaign can optimize toward the configured action while still charging for valid clicks.
For a lead test, choose the objective only after checking measurement readiness. A click-focused test can help examine traffic and landing-page behavior, but you should still measure accepted leads. An impression-focused test answers an exposure question; judging it solely on immediate demo requests would mix two different briefs.
A conversion objective requires an eligible active standard conversion event setting from the campaign's account. Do not assume a custom “qualified opportunity” event can automatically become the optimization goal. Official conversion tracking guide.
There is no useful universal CPC, CPM, or minimum test budget for every US business in this article. Set financial boundaries from your margin, capacity, acceptable acquisition cost, and the question you need answered. Documentation sample bids are examples, not market-price forecasts.
Use context hints to describe buying situations
Context hints are signals about relevant conversations, not an exact-keyword buying system. They also do not replace geographic targeting. Official targeting guide.
For the HVAC example, possible planning hints might describe homeowners comparing replacement options, evaluating installation estimates, or deciding whether to repair an aging unit. These are illustrative ideas, not a promise that a particular sentence triggers an ad.
For scheduling software, describe the operational situation: managers coordinating shifts across locations, reducing manual scheduling work, or comparing systems with an existing payroll process. Avoid dumping every synonym into the field. A coherent problem is more useful than a bag of loosely related marketing terms.
Treat geography separately. A Phoenix reference in a hint is not a geographic fence. Inspect the supported location controls and select the actual coverage available in the account. The landing page should also state the business's service territory. Neither step excuses the other.
Review lead fit after launch. If the software campaign brings individual freelancers while your product requires multi-location teams, the issue may be the problem framing, the offer, the landing page, or several of them. Write down the mismatch before changing settings.
Build creative that earns the next step
The documented standard chat-card format contains a title, body, image, and landing-page URL. Titles must be 3–50 characters; bodies cannot exceed 100 characters. Images can be JPEG, PNG, or WebP and must be at least 640 × 640 pixels. Official creative requirements.
Those constraints favor specificity. You do not have room for a company biography.
Illustrative local-service copy:
- Title: “Compare AC Replacement Options”
- Body: “See installation options and request an estimate for your home.”
Illustrative B2B software copy:
- Title: “Scheduling for Multi-Location Teams”
- Body: “Explore shift planning across locations. Request a product demo.”
Use only offers the advertiser actually provides. Do not add “free,” “same day,” financing terms, integrations, or service guarantees without approval and evidence. A fictional example is not permission to reuse its promises in a real account.
Choose an image that supports the decision: a genuine installation photo, a relevant product view, or a clearly identified illustration. Do not fabricate a customer, certification, dashboard result, or endorsement. Check the crop at the size where it will be seen; tiny interface text rarely explains a software product well.
Test two materially different messages, not two punctuation changes. One software variation might emphasize multi-location coordination; another might emphasize replacing spreadsheet scheduling. Keep the offer and destination consistent while testing that difference.
Match the landing page to the ad
The ad opens a question. The landing page should answer it immediately.
For the HVAC example, explain the service, coverage, what an estimate includes, and the next step. If the company cannot handle emergency work, do not let the page imply that it can. Use real project images and support any licensing or credential claims with current evidence.
For software, show the relevant workflow, who the product suits, important limitations, and what happens in the demo. If a buyer must meet a company-size requirement or needs a specific integration, surface that before asking for contact details.
Use one primary action. A demo page should not make the visitor choose between a newsletter, an unrelated download, a general contact form, and a sales calendar with equal emphasis. Secondary information can support the decision without competing with it.
Test the complete mobile experience: page load, image crop, text readability, keyboard behavior, error messages, submission, and confirmation. A form that looks clean but loses the source parameters is not ready. A confirmation page that loads despite a failed submission is worse.
If the landing page needs work, fix the affected journey before expanding traffic. Our web design work should serve that journey, not decorate a weak offer.
Track successful submissions, not button clicks
Define “lead created” as a completed action in your system. The visitor has submitted valid information, the server has accepted it, and a lead record or equivalent durable record exists.
A submit-button click happens before you know any of that. The form may fail validation, lose its connection, hit a server error, or be rejected as spam. Counting the click as the lead makes failures look like success.
Have the application produce a stable lead or submission ID after acceptance. Trigger the measurement event from that confirmed state. OpenAI's tracking guide includes lead_created as a standard event for lead submissions. Official conversion tracking guide.
Separate at least these business stages: accepted inquiry, qualified lead, booked appointment or demo, attended appointment, proposal, and won customer. Your advertising event does not have to contain the entire sales process, but your review must not stop at its first step.
Before launch, run controlled tests for a successful submission, invalid form, failed server response, repeated click, and confirmation-page refresh. Confirm which actions create a record and which emit an event. Remove or label test records in reporting so QA does not become apparent campaign performance.
For telephone-heavy services, distinguish a click on the phone number from an actual connected call and from a qualified call. If you cannot measure the latter stages, report that limitation rather than silently upgrading a click into a lead.
Preserve UTMs and the ad-click reference
Use campaign and ad identifiers to connect visits with creative. The documented query template supports {campaign_id}, {ad_id}, and {oppref}. Account defaults can be overridden by campaign, ad-group, and ad settings when the same parameter key is configured. Official tracking parameter guide.
Keep the naming scheme consistent with your analytics system. A practical plan is a stable source and paid medium, the campaign ID in the campaign field, the ad ID in the content field, and a dedicated field for the click reference. Test the resolved URL; do not assume a template has produced the expected values.
Capture the attribution fields at arrival and preserve them through the form, redirects, and any separate booking domain where permitted. Add them to the lead record rather than leaving them only in a browser session.
The Conversions API does not collect oppref automatically: your integration must preserve the original reference and supply it on the event. Official Conversions API guide.
Do not put names, email addresses, or private form answers into UTMs. Tracking parameters are for attribution, not for exporting the customer's inquiry into URLs. Decide how consent, storage, and permitted matching data will be handled before implementation.
Our web analytics service focuses on making that chain inspectable: the visit, the accepted action, and the record the sales team actually receives.
Deduplicate Pixel and Conversions API events
The Measurement Pixel sends events from the browser. The Conversions API sends them from your server. Using both requires coordination, not two unrelated implementations. Official Measurement Pixel guide.
For one action sent through both paths, use the same Pixel ID, event type, and event identifier: Pixel event_id must match API id. Reuse the identifier on retries of the same action. Keep the API key on the server. A validate_only request checks the payload but does not save an event. Official Conversions API guide.
The application should generate the identifier once. If the browser invents one ID and the server invents another, you have described two actions instead of one. If every retry gets a new ID, a temporary network failure can inflate reported performance.
Validate the payload first, then run a controlled saved-event test and inspect the intended source. Check your internal record and event monitoring together. Acceptance by an endpoint is not evidence of an attributed customer; event delivery, attribution, qualification, and revenue are different checks.
Keep failure handling visible. Log safe diagnostic details, preserve the action ID, and investigate rejected events. Do not repeatedly resend a whole set of leads without knowing which actions were already accepted.
Judge economics using qualified leads
Here is an illustrative USD example, not a benchmark, quote, or campaign result.
Suppose a test spends $1,200 and produces 30 accepted inquiries. The raw cost per inquiry is $40. Sales then identifies 12 qualified inquiries, making the cost per qualified inquiry $100. Three become customers, giving an advertising-only acquisition cost of $400 per customer.
If each customer produces $1,000 in contribution before acquisition costs, those three customers represent $3,000 in contribution. Subtracting $1,200 in ad spend leaves $1,800 before management, landing-page work, sales effort, and other relevant costs. That is not the same as $1,800 in net profit.
Now suppose the same spend produces 60 inquiries but only six qualified prospects and one customer. The raw inquiry cost improves to $20, while the qualified-inquiry cost worsens to $200 and advertising-only acquisition cost becomes $1,200. A prettier lead dashboard can describe a worse business outcome.
For software, account for sales-cycle lag, churn risk, delivery cost, and cash timing. Do not treat unsigned pipeline as revenue or use optimistic lifetime value to justify any acquisition cost. For local services, account for travel, appointment capacity, job margin, and seasonal demand.
Agree on qualification before reviewing results. Record rejection reasons such as outside coverage, unsupported need, spam, duplicate, or unsuitable company profile. These categories turn “the leads are bad” into something you can improve.
Troubleshoot delivery before raising spend
If impressions are absent, work through the serving prerequisites before changing the bid. Inspect the account, campaign, ad group, ad, reviews, schedule, targeting, budget, and reported serving issues. An active child resource does not bypass its parent campaign's state. Official campaign management guide.
If impressions arrive without useful clicks, inspect the offer and creative. If clicks arrive without accepted leads, test the landing page and form with the actual destination parameters. If your system has leads but the platform has no events, investigate measurement and attribution rather than rewriting the offer first.
If reported events exceed your accepted records, check duplicate submission behavior, refresh events, and Pixel/API identifiers. If accepted leads are plentiful but qualification is weak, inspect buyer fit and page clarity. If qualified prospects fail to become customers, review sales follow-up, response time, pricing, and delivery constraints.
Keep reporting comparisons aligned: date range, timezone, attribution settings, and definition of conversion. Different tools may assign credit differently. Reconcile the discrepancy; do not pick whichever number makes the test look better.
ChatGPT Ads, Google Ads, and organic GEO do different jobs
Do not move an entire acquisition budget because a new channel exists. Compare channels against the same accepted-lead and commercial definitions, while respecting the different intent and placement contexts.
A local company with an urgent-service offer should evaluate its current search acquisition separately from a conversational research offer. A software buyer comparing operational approaches may need education before a demo. Those are planning hypotheses to test, not guaranteed audience behaviors.
Our Google Ads management work can provide a comparison framework: use equivalent geography, qualification rules, and fully loaded costs where possible. Keep channel-specific spending and outcomes visible instead of blending everything into one average.
Organic GEO is separate again. It concerns whether useful, credible business information can be found and represented in generated answers. Evaluate paid placement through campaign reporting and organic discovery through separate observations. We would never describe purchasing an ad as a guarantee of organic inclusion, or present a sponsored placement as evidence of an organic recommendation.
Our generative engine optimization service addresses the organic work. Track it independently. An ad campaign report cannot prove an improvement in organic AI visibility.
Launch checklist and decision rules
Before spending, verify the following against the actual account and website:
- The advertiser, access, currency, timezone, and relevant eligibility are checked.
- One buyer, offer, geography, objective, and financial boundary are documented.
- Creative uses supported specifications and substantiated claims.
- The destination matches the ad and works on desktop and mobile.
- A successful form creates the expected lead record; invalid submissions do not.
- Source fields and the click reference survive the conversion journey.
- Browser and server events share identifiers where deduplication is needed.
- A saved test event is observed, not only a validation response.
- Sales owns follow-up and records qualification and rejection reasons.
- Someone can pause the campaign and explain when they must do so.
Pause immediately for a broken form, unintended spend exposure, incorrect territory, or misleading offer. Repair the defect before resuming.
Continue a bounded test when measurement is reliable but the sales cycle has not matured enough to judge customers. Do not call early pipeline a win. Change one meaningful variable when you have evidence of a specific mismatch.
Consider scaling only when accepted leads reconcile with your records, quality is acceptable, economics meet the agreed threshold, and the business can handle additional demand. Increase exposure deliberately and watch the same quality measures again. More volume can change the mix.
If those conditions never hold, stopping is a valid decision. The purpose of a test is to learn whether this channel belongs in your acquisition system, not to defend the launch.
Frequently asked questions
How do I start advertising on ChatGPT in the US?
Check your ad account's access and eligibility, then define one audience, offer, landing page, measurement event, and spending limit. Build the campaign, ad group, and ad in a paused state; verify review, tracking, and a successful lead submission before activating delivery.
Does ChatGPT Ads support conversion campaigns?
Yes. Current documentation includes impression, click, and conversion objectives. A conversion campaign requires an eligible active standard event setting from the same account. It optimizes toward that action but is billed for valid clicks, not a guaranteed price per lead.
Are ChatGPT context hints the same as Google Ads keywords?
No. Context hints describe relevant conversational situations. They are not exact-match keywords and do not replace geographic targeting. Use them to explain the buyer's problem, then separately configure the supported location settings for your service area.
What should count as a lead conversion?
Count the successful creation of a lead after the form has passed validation and your system has accepted the submission. A submit-button click, form opening, or failed request is not a lead. Evaluate qualification and sales outcomes separately in your CRM.
How do I avoid counting Pixel and API conversions twice?
Send the same action with the same Pixel ID and event type, using the Pixel event_id as the Conversions API id. Preserve that ID when retrying the event. Verify a test submission in event monitoring and in your own lead records.
Does buying ChatGPT ads improve organic AI recommendations?
No paid placement should be sold as a way to buy an organic recommendation. Sponsored advertising and organic answer visibility are separate activities. Evaluate ads through their campaign and business outcomes; evaluate GEO through independently observed organic visibility.
Want help defining the campaign and measurement chain? Contact YAG with your offer, service area or buyer profile, landing page, and current lead workflow. We can scope the next step without promising a CPC, a lead price, or an organic recommendation we cannot guarantee.