How to run ads on ChatGPT after launch.

Launch day: prove the system works

Before interpreting performance, verify that a real homeowner can travel from ad click to a correctly handled inquiry.

1. 1

   Inspect live delivery

   Confirm the approved offer, service, destination URL, territory, schedule, budget limits, and available platform settings match the launch sheet.

2. 2

   Test every conversion path

   Submit forms on mobile and desktop, tap phone links, place a test call, inspect confirmation states, and verify required consent and privacy language.

3. 3

   Follow the data

   Confirm each test event appears in analytics and platform reporting where supported, carries the correct source, and creates or updates the expected CRM record.

4. 4

   Follow the lead

   Make sure forms reach the assigned people, calls ring the intended queue, notifications arrive, and after-hours routing behaves as documented.

5. 5

   Confirm human coverage

   Tell intake and sales the offer is live, provide qualification rules, identify the escalation owner, and verify phones are staffed for the advertised schedule.

The first 14 days: observe without freezing

Early data deserves attention, but a handful of events rarely supports sweeping conclusions.

Fourteen days is an operating window, not a universal statistical rule. A material tracking fault should be fixed immediately; a compliance concern should be paused immediately. Conversely, low activity alone may reflect limited inventory, account constraints, narrow eligibility, or normal variation. Document what is known and what remains uncertain before acting.

Use the period to establish a clean baseline. Annotate outages, holidays, storms, capacity changes, and offer interruptions. Without that context, the team may later credit or blame campaign edits for changes caused by the business itself.

Run one disciplined weekly cycle

A regular review prevents both neglect and constant tinkering.

1. 1

   Reconcile delivery

   Compare actual spend with the weekly plan, confirm campaign status and eligibility, and flag material under- or over-delivery.

2. 2

   Audit the funnel

   Reconcile reported conversions with call records, form receipts, CRM entries, duplicates, spam, and missing source data.

3. 3

   Review lead quality

   Group inquiries by qualification and loss reason, then inspect the underlying conversations rather than relying only on totals.

4. 4

   Check operations

   Review answer rate, speed to first attempt, follow-up completion, appointment handling, territory declines, and current crew capacity.

5. 5

   Choose one change category

   Select creative, context and eligibility inputs, geography, landing page, routing, or budget. Avoid changing several categories unless correcting an urgent fault.

6. 6

   Record the decision

   Document the evidence, hypothesis, exact change, start date, expected signal, review date, and person responsible.

Know what you are allowed to change

The weekly rule is not “one tiny edit.” It is one coherent hypothesis. If repeated inquiries are outside the service area, a geography correction and matching copy clarification belong together. If qualified visitors abandon a long form, a landing-page test may include field reduction and revised supporting text. The change should answer one operational question.

Do not combine a new offer, larger territory, higher budget, different page, and revised intake script, then attribute the outcome to creative. That creates a new campaign with no useful comparison. Exceptions are failures that harm customers or violate approved limits: broken pages, incorrect claims, misrouting, policy issues, and uncontrolled spend should be addressed at once.

Keep a change log accessible to marketing, sales, and leadership. Memory is unreliable when several people can edit accounts or websites. The log makes reversals safer and stops the same failed idea from being rediscovered a month later.

Manage pacing without chasing the meter

Budget control is a weekly forecast tied to capacity, not an emotional response to each day’s spend.

Reforecast remaining spend using the actual run rate and known schedule, then compare it with the cap. If the platform’s controls or reporting behave differently than expected, state that uncertainty rather than assuming precise pacing is possible. Never spend into a fulfillment bottleneck just because budget remains.

Build the lead-quality feedback loop

Every inquiry needs a source, owner, first-response timestamp, qualification status, appointment status, estimate outcome, sale outcome where available, and a concise loss reason. Marketing does not need private sales commentary; it needs consistent operational facts. Sales needs a short workflow that fits the CRM rather than a spreadsheet nobody maintains.

Hold a brief weekly review with one person empowered from each side. Read representative records: a qualified lead that sold, one that did not, a disqualified inquiry, and an unreachable contact. Ask whether the ad promise matched the conversation, whether the location and service were valid, whether response was timely, and whether the stated loss reason is supported.

Use count only as the top of the story. Ten inquiries with unknown dispositions reveal less than a smaller group with verified service fit, appointments, and outcomes. A low count can still expose useful message or routing issues; a high count can conceal spam, out-of-area demand, or calls abandoned on hold. Optimize toward qualified opportunities and economic outcomes as the data matures.

Pause, hold, or scale for a documented reason

Scaling is not a reward for a good-looking dashboard. Confirm that qualified opportunities become appointments and sold work at economics the business accepts, allowing for the sales-cycle delay. Increase in controlled steps and keep watching quality and capacity. Prior results do not guarantee that additional spend will behave the same way.

• Pause

  Stop for compliance concerns, materially inaccurate claims, broken conversion paths, spend beyond authorization, persistent misrouting, no fulfillment capacity, or evidence that the offer cannot be delivered.

• Hold

  Maintain the current setup when systems work, spend remains controlled, evidence is still thin, lead outcomes are pending, or a recent test has not had a fair observation window.

• Scale carefully

  Consider incremental expansion only when verified lead quality and downstream economics support it, operations have capacity, tracking is trusted, and the next budget step remains affordable.

• Revise

  Make a bounded change when repeated evidence identifies a correctable mismatch in message, territory, destination, routing, or campaign inputs.

Avoid the mistakes that kill a promising test

The familiar failure pattern is operational: the team changes multiple variables after a quiet day, judges performance before sales outcomes mature, misses calls during advertised hours, and follows forms slowly. Marketing then optimizes against incomplete data while sales concludes that the source is poor. None of this proves the channel could or could not work; it proves the test was not controlled.

Assign owners before problems occur. One person protects budget and change control, one owns tracking and routing, and one owns lead disposition and follow-up. Review unanswered calls and aging leads as urgently as ad settings. A relevant placement cannot compensate for a ringing phone nobody answers.

Finally, do not let novelty lower the standard applied to other paid media. Require accurate claims, documented access and platform limitations, sensible economics, clean measurement, and honest reporting. The channel is still developing, so disciplined operations (and comfort with uncertainty) matter more, not less.