Which Ad Copy Attracts Leads That Close

Below: the two rates, getting stages per ad, wording hypotheses, the test design, the arithmetic, and how to read the result.

Cost per lead rewards copy that removes doubt, and doubt is often what filters out the wrong people. "Free guide, no strings attached" converts well because it asks for nothing. "For B2B teams spending over $20k a month on ads" converts worse because most readers are not that team, which is exactly what you want them to learn before they fill the form.

Ads Manager calls A the winner at half the CPL. The CRM says B delivers more qualified leads for the same money, and sales spent their week on 100 conversations instead of 200. More on why the platform's number drifts from yours in the true cost per lead.

Qualification rate is fast: your team decides within days, and the counts are large. Close rate is slow and small, but it is where the money is, and copy that over-promises can qualify well and close badly. Track both, plus cost per qualified lead and revenue per ad, which fold spend back in. What "qualified" means is your definition, set once; see qualified lead.

None of this works unless every lead in the CRM knows which ad it came from, and its stage is kept up to date. Three things have to be in place before the first copy test.

1. The ad on the lead. Either the ad id in a UTM parameter carried into a hidden form field, or the click id (fbclid, gclid, li_fat_id and the others) kept with the visitor and resolved to the ad. The click id survives a return visit; a URL parameter often does not. See click id.
2. Stages that mean the same thing for every rep. A written rule for qualified, and the closed-won stage used consistently. A stage one rep sets on the first call and another after the third makes the qualification rate a measure of the rep.
3. One place where both meet. Spend per ad from the platform and stages per ad from the CRM, joined by the ad. A spreadsheet works for one test; it does not survive ten.

No study we can cite tells you which wording brings better leads in your market, and a pattern that works for one offer fails for the next. These are hypotheses worth a test, each with the trade it probably makes. Your CRM decides.

• Name the buyerCalling out a role or company type in the first line ("For agency owners...") may cut volume and raise qualification rate, because the wrong readers self-select out.
• Mention the price or a minimum"From $X a month" or "for teams over N people" may cost leads and save sales calls. Watch close rate: price-aware leads may close faster.
• State the real next step"Book a 30-minute demo" sets a higher bar than "Learn more". Fewer leads, possibly more serious ones. The button is part of the copy.
• A specific problem over a general benefit"Your CRM says Direct for 40% of deals" versus "Better attribution". Specific pain may attract the people who have it.
• Proof that matches the buyerA named customer like the reader may qualify better than a big number with no context. Test the proof type, not only the wording.
• Remove the baitWords like "free", "instant" and "no commitment" may raise volume and lower qualification rate. Worth testing by taking them out.

The figures inside these examples (the 40%, the $X, the N people) are placeholders. Use your own, and only claims you can back. To find which patterns your past ads already used, tag them with the template in AI ad creative analysis.

• One change per testChange the hook, or the ask, or the qualifier, not all three. Otherwise a result cannot tell you which change did it.
• Same audience, same budgetMeta's A/B test feature (a tool group in its ads MCP server too) is built for this. Two ads in one ad set are quicker to set up, but delivery decides how much each gets, so spend rarely splits evenly. Compare rates, not totals, either way.
• The metric decided up frontQualification rate as the main read, close rate and cost per qualified lead as checks. Write it down before launch.
• A run length tied to your cycleStop spending when each version has the leads you planned, then wait until your typical time to qualify (and later, to close) has passed before reading.
• Hands off while it runsNo budget shifts, no audience edits. Each change starts a new test inside the old one.

There is no magic number of deals, and we will not invent one. What we can do is show the arithmetic, so you can run it on your own counts. It uses the standard error of a proportion, √(p × (1 − p) ÷ n), and the rough 95% range of p ± 2 standard errors. With small counts this approximation is crude, which only strengthens the point.

The ranges overlap almost entirely. Version A looks two thirds better, and the data cannot tell you whether it is better at all. Seen as counts: a count of 5 has a natural spread of about √5 ≈ 2.2, so a version that "really" closes 5 in 25 will often show 3 or 7 by chance alone.

How many it would take

To see a gap of 8 points (20% against 12%) at about two standard errors, the standard error of the difference, √((0.20 × 0.80 + 0.12 × 0.88) ÷ n) = √(0.2656 ÷ n), has to fall to 4 points. That gives n ≈ 0.2656 ÷ 0.0016 ≈ 166 qualified leads per version, or roughly 33 and 20 deals. And that is only the level where the gap stops looking like chance; it is not a guarantee that a real gap will show.

The same arithmetic on qualification rate is kinder, because leads are plentiful. With 200 leads per version qualifying at 25% and 15%, the standard error of the difference is √((0.1875 + 0.1275) ÷ 200) ≈ 4 points, and the 10 point gap is about 2.5 standard errors. That is why qualification rate is the main read and close rate the check.

An AI client with two connectors does the joining in a chat. The Meta ads MCP server reports spend and results per ad, and its tool groups include A/B tests; the LeadJourney MCP server answers with leads, qualified leads, won deals and revenue per ad from the CRM. Prompts that work:

• "For the two ads in this test, give me spend, leads, qualified leads and won deals since launch, with qualification rate, close rate and cost per qualified lead."
• "Compute the standard error for each rate and say whether the gap is larger than two standard errors."
• "Quote the first line and the button of each ad next to its qualification rate."
• "Across the last 90 days, which ads have the highest qualification rate with at least 50 leads? What do their first lines have in common?"

Claude proposes; you decide. Writing a new version or pausing the loser goes through the Meta connector, where new ads land paused, or through Ads Manager. More in Meta Ads MCP plus your CRM and Claude for Meta Ads.

LeadJourney does not write or test ad copy and does not change your ads. It makes the per ad lead quality readable. One script, live in about 21 minutes, tracks every click, form fill and call server-side and first-party at 95%+ accuracy, keeps the click ids and joins each lead to its CRM deal in HubSpot, Salesforce, Pipedrive, Close, Attio, GoHighLevel, ActiveCampaign, Odoo, Zoho CRM or Microsoft Dynamics 365, with the stage it reaches.

So every campaign, ad set and ad on Meta, Google, LinkedIn, Microsoft, Reddit, X, Taboola and Outbrain shows leads, qualified leads, won deals and revenue beside its spend. The stages you map go back to the platforms as conversions, and the MCP server lets Claude or ChatGPT read the same numbers. It reads; it never edits an ad or a CRM record.

• AI ad creative analysis: hook, body and close, judged on revenue
• Automated budget rules on cost per qualified lead
• Why your ad platform's MCP gives Claude the wrong CPL
• Meta Ads MCP plus your CRM: which campaigns made revenue
• Claude for Meta Ads: connector, skills and 25 prompts
• How to increase the quality of your leads in performance marketing
• Fix Meta Ads lead quality
• Lead quality per channel and campaign