$1,000 a month, six weeks, one lead-gen business: that is the entire ChatGPT Ads test, media spend included, and I am not letting it creep. I would run it beside your Google Ads, not as a separate campaign whose dashboard gets to declare its own victory.

Use the same conversion event, landing page logic and lead-qualification rule on both sides. Then judge ChatGPT on cost per qualified lead against Google. CTR and reach are how a weak test stays alive after you should have killed it.

Here is the plan in order. The weekly pace is roughly $225, putting six weeks at about $1,350: within the $1,000 monthly cap, without a top-up. I cut the extras agencies like to sell with a new channel: broad awareness themes, a separate landing page build and creative testing. With roughly 200 to 333 clicks available from a $1,000 month at the starting CPCs cited below, those extras consume the test instead of improving it.

1. Week 0: Decide whether to run it. Spend $0, avoid an unreadable test

Before setting up a card, check the base account. If Google Ads does not already produce qualified leads you can measure, skip the sidecar. A second channel cannot repair a broken conversion definition. It gives you another set of numbers to argue about.

Skip it, too, if nobody can grade calls or forms for six weeks. You need a person to mark whether a lead was accepted, not someone to admire the impressions chart. And think about who can see the ad: logged-in adults on Free and Go may see ads, while Plus, Pro, Business, Enterprise and Education users do not. If the buyers you need are on paid ChatGPT plans, this test cannot reach them there.

My short skip list is:

  • No steady Google baseline: fix conversion tracking before buying sidecar traffic.
  • Average job value under $500 with a one-call close: you need volume, not a roughly 200-click experiment.
  • Paid-plan buyers only, or EEA-heavy demand where only current-conversation matching runs: reach is too thin to learn much at this budget.
  • No one grading lead outcomes: you will fall back to CTR, the metric this playbook exists to ignore.

This screening costs nothing against the media cap. It buys you the right to treat the result as a decision rather than a story about why the dashboard looked interesting.

2. Week 0: Build the comparison. Spend $0 in media, buy a fair read

At $3 to $5 starting CPC, with a $3.00 default and a $25-per-day campaign minimum in the US, $1,000 buys about 200 to 333 clicks before a single conversion. Every click needs to land in a comparison you can trust.

The ChatGPT unit is one compact sponsored card below the answer, with a title, description, square image and one link. It is not a Google search ad in a different color. But the measurement needs to meet the same standard: use the same conversion event, consistent UTM naming with utm_source=chatgpt beside utm_source=google, and CRM send-back through pixel plus Conversions API before launch. Send both channels through the same landing page logic and the same follow-up.

I used to let clients launch with platform-reported leads and sort out lead quality later. I was wrong. If you cannot reconcile a lead to a CRM outcome, you cannot tell whether one channel bought customers or merely bought form fills. If call tracking is in play, keep the __oppref ID on every call so both channels face the same disqualification logic.

Write the stop and scale rules now, in numbers, and get whoever runs the account to agree. My default stop rule is kill the ChatGPT sidecar at $500 spent with zero sales-accepted leads, provided Google is still producing at its normal cost per qualified lead. Do not promise to wait until Week 4 if you hit that condition in Week 3. Independent tests cited here cluster around $2 to $5 CPCs with conversion rates near 0.21%, against 3.71% on Google; one spent $2,319 for zero MQLs while Google produced two. At $3 CPC and a 0.21% conversion rate, the arithmetic is about $1,429 per conversion. The stop rule exists to keep you from paying that bill just to confirm the arithmetic.

The scale rule is simpler: no budget increase without your written approval, even if early CTR looks good. Guides that lead with CTR and reach give a weak lead-gen test room to look alive. I have watched a 1% CTR keep a dead test funded for two extra months. CTR does not pay for callbacks.

If an agency runs this, ask for its test plan before you pay: starting hints, bid, daily cap, dollar-and-lead stop rule, and the person who approves an increase. I run this work inside groas Search Ads, which covers Google, Bing and ChatGPT under one flat fee rather than rewarding more sidecar spend. Bid, hint and budget changes get logged with their reasons; scaling still needs your sign-off. Buy that discipline, not a prettier CTR report.

3. Week 1: Launch one narrow campaign. Spend about $225, buy first delivery

Start with one campaign, one ad group and five to eight context hints written as buyer moments, not keywords. A hint might describe a homeowner comparing quotes for emergency drain repair this week, rather than simply saying plumber near me. Hints are signals, not exact-match rules. You do not get a search-terms report or a view of which prompt fired the ad.

That limited visibility is the reason to stay narrow, not an invitation to add every possible buyer. In a relevance-weighted second-price auction with one winner per conversation, a relevant hint and page can beat a higher bid and pay less than the maximum. Ten loose hints spread a small click budget across intents you cannot inspect. One coherent buyer problem buys a cleaner read.

Set the campaign at $30 to $35 per day. Daily budgets work as seven-day averages, with a day able to spend up to twice its cap within the weekly limit, so watch the week rather than treating each day as a separate invoice. Start at a $3.50 max CPC and leave it alone for seven days. If delivery is still thin and the cap is going unspent after that week, raise the bid once to $5 before widening the hints.

Cut the second ad group, second creative and image variants. The median ad cited here uses about 26 title characters and 57 description characters, and 95% include an image. Ship one card with one square image. Send it to the page already used for the comparable Google offer. A dedicated ChatGPT page may feel like progress, but on this budget it adds a landing-page variable when you are trying to test a channel.

Do not mistake thin delivery for a creative problem, either. The paid-plan exclusion may mean the person you most want never sees the card. Week 1 buys an initial delivery read, not permission to rebuild the campaign every morning.

4. Weeks 2–3: Grade the leads. Spend about $450, buy a quality read

If you are still running by the end of Week 3, you will have spent roughly $650 to $700 total. At the starting CPCs, that is roughly 150 to 200 clicks. Open the CRM before you open the CTR column. How many clicks became sales-accepted leads under the rule you use for Google? Did the person answer the phone, match the service area and have a workable budget or timeline? Apply your actual rule word for word to both channels.

In one same-advertiser test, Workshop Digital spent $2,319 at $3.14 CPC for a 1.06% CTR and zero MQLs, while Google Search at 5.2% CTR produced two. A report organized around clicks could keep that ChatGPT campaign in the meeting. A report organized around qualified leads gives it nowhere to hide.

Cheap clicks create the same trap. In one B2B comparison, ChatGPT clicks cost $0.76 against $10.46 on Google, but converted at 0.8% against 6.64%, producing a $92.21 cost per lead against $157.57. Raw cost per lead appears to favor ChatGPT. Yet Google leads became paying customers at a materially higher rate in that comparison window. That is why this playbook does not award the win to the cheaper form fill. A $92 lead that never books is more expensive than a $157 lead that does.

Grade each lead without the channel label if you can. Use the same listener and checklist for calls, the same rule for forms, and the same follow-up. If the person needs what you sell, in the area you serve, on a timeline you can work with, count the lead. Otherwise, do not. If ChatGPT reaches $500 spent with zero accepted leads while Google keeps qualifying normally, stop now. You do not owe Week 4 a ceremony.

5. Week 4: Make the midpoint cut. Spend about $225 only if the test survives

At the planned pace, total media spend approaches $900 by the end of Week 4. Divide ChatGPT spend by its sales-accepted leads and place that cost per qualified lead beside Google’s for the same four weeks. If there are no accepted ChatGPT leads, the $500 rule should already have ended the test. A pause is not a failed experiment; paying past a written stop rule is.

If the sidecar has fewer than three qualified leads and sits at more than double Google’s cost per qualified lead, pause it. If it has a plausible quality signal but not enough to decide, keep the campaign at the same cap through the final two weeks. Do not claim to know which individual hint produced a callback: the setup does not give you that prompt-level view. Cutting a supposed losing hint on the strength of a dashboard guess would turn the midpoint review into another creative test you cannot afford.

The cut here is spending without a quality signal. What remains buys one last, bounded chance to see whether the lead pattern holds.

6. Weeks 5–6: Decide. Spend up to about $450 more, buy an answer

Six weeks at this pace is about $1,350 in total media, inside two $1,000 months without a top-up. This is when owners ask whether another month might make the channel look better. Maybe. But OpenAI publishes no cross-advertiser benchmarks and describes delivery as a beta where spend can fluctuate. More time without qualified leads buys more fluctuation, not a stronger argument.

Make one of three calls against Google’s cost per qualified lead:

  1. Kill if ChatGPT delivered zero or one qualified lead while Google delivered at its normal rate, or if the completed test remains more than 50% above Google’s cost per qualified lead. Pause it, keep the tracking in place and write the reason in one line.
  2. Hold if ChatGPT delivered at least two qualified leads within 50% above Google’s cost per qualified lead but has not earned a scale decision. Keep the cap where it is for one more month. Change nothing except a weak hint you can justify from what qualified callers actually needed, not a guess about which prompt fired an ad.
  3. Ask for approval to double only if ChatGPT matched or beat Google on cost per qualified lead across at least five qualified leads. Approval comes before the budget move. A card that gets clicks but no callbacks does not earn a second $1,000.

This is capped-test logic: learn whether the channel produces the outcome you buy ads for, then scale only what converts. Reach does not get a vote.

When the constraint lifts, keep the same rule. If five or more qualified leads put ChatGPT at or below Google’s cost per qualified lead, ask to move it to $2,000 and add a second ad group around the buyer need that surfaced in accepted calls. If it is within 50% above Google but has not earned that decision, keep the $1,000 monthly cap and keep grading leads. If you killed it, leave the pixel in place and retest in two quarters. The budget stays small until the leads say otherwise.