How to prove to a client that a page change helped

By Ranjit Bhinge, founder of Blur Studio · Updated · 11 min read

Short answer: You prove a page change helped by showing the old page and the new page ran at the same time, to similar visitors, with one goal counted the same way, over enough visits that luck is unlikely to explain the gap. If you cannot run them side by side, a careful before and after is weaker proof, and you should say so. Tools such as Unbounce, a landing page builder, or Ploy, an AI platform that builds and runs business websites, can split the traffic for you, but the report you hand the client matters as much as the tool.

Ranjit Bhinge

The client asks, "Did the new page actually work?" A screenshot of a bigger number is not an answer. Here is the evidence a sensible client will accept.

I'm Ranjit, and I run Blur Studio: 15+ years building software, 120 apps taken live, and clients who have gone on to raise $10M+. I build AI systems and processes that help businesses do a lot more with a lot less time. I also work differently from most of what else is out there, more on that further down this page. Say hello.

To prove a page change helped, show four things on one page: what you changed, how long and to how many visitors both versions ran, what the one agreed goal did on each, and what else could explain the difference. Best proof is a split test, where the old and new page run at the same time. Second best is a before and after with the other changes listed. Anything weaker is an opinion, and you should call it that.

This guide is about the evidence and the wording. For launching version B safely, read how to try a new version of a weak ad landing page safely. For keeping bookings safe while a test runs, read how to run a website test without losing bookings. For linking leads to money, read client reporting that ties leads to revenue. Prices and features here are as of November 2026, so check each vendor page before you pay.

This page may contain affiliate links from which Blur Studio earns commissions.

What counts as proof, from strongest to weakest

LevelWhat it isWhat you can fairly say
1. Split test, planned in advanceOld and new page run at the same time, visitors assigned by chance, sample and goal set before starting"The new page produced more enquiries than the old one in a fair test."
2. Split test, planned afterwardsBoth pages ran together but nobody wrote down a sample or stop rule"The new page did better in this period. We could be seeing luck."
3. Before and after, nothing else changedSame ads, same budget, same weeks of the year, one page change"Enquiries rose after the change. It is consistent with the page helping."
4. Before and after, other things changedNew ads, a sale, a season, or a price change happened too"We cannot separate the page from the other changes."
5. A bigger number on a dashboardNo comparison period or sample shownNothing yet.

Aim for level 1. Settle for level 3 when traffic is too low for a split, and label it plainly. A client who is told the truth about level 3 trusts you more the next time, when you can show level 1.

Agree the question before you change anything

Most arguments about proof start because nobody agreed the goal. Put these four lines in an email to the client before the change goes live:

  1. The goal: one action, for example "a form sent or a booked call". If you count phone taps and form sends together, say so.
  2. The change: one sentence on what is different on page B.
  3. The sample and time: for example "at least 1,500 visitors to each version and two full weeks", with the visitor number worked out from your own rates (see below).
  4. The decision: "If B beats A at the sample, we keep B. If not, we keep A."

When the client has signed off on the rules, the result cannot be argued with afterwards. It also protects you. If B loses, you are reporting a test that worked as intended, not a failure.

The one-page evidence pack

This is the report to hand over. Keep it to one page. Evidence goes first, story second.

  1. Headline sentence. "Version B produced 5.7 percent enquiries against 4.0 percent for version A over 3 weeks and about 350 visitors each, well short of the 2,500 or so we planned. Result: promising, not conclusive." Use the plain verdict from the table above.
  2. What changed. A screenshot of each version with the one difference marked.
  3. How it was run. Dates, split (for example 50/50), the goal, the sample set in advance, and whether both ran at the same time.
  4. The numbers. A small table: visitors, enquiries, rate, cost per enquiry, and (where you have it) enquiries that became jobs.
  5. What else changed. Ads, budget, season, price, site speed, outages. If nothing, say "nothing else changed" and name who checked.
  6. Limits. One short paragraph on sample size and what the numbers cannot tell you.
  7. What we do next. Keep B, run another round, or revert, and the date.

If you already send monthly reports, add this as one section of that report. The agency client reporting template shows where a page-test section fits, and how to explain ad results to a client who asks if ads are working helps with the conversation.

Count it fairly: five checks before you show a number

  1. Same goal, same way. Count form sends and calls on both versions in the same way. A form that sends a confirmation page on A but not on B will make A look better for no reason.
  2. Same period. Compare A and B across the same days. Do not compare B in March with A in February.
  3. Same traffic. Both versions should get clicks from the same ads. If B only gets the cheaper ad group, the comparison is not fair.
  4. Whole business cycles. Run for at least one to two full cycles. Optimizely, a large testing company, says at least one full cycle, typically seven days. Convert, another, says one to two full business cycles. The two differ, so state which rule you used.
  5. Real leads, not just form sends. A new page that gets more form sends from people who never answer the phone is a loss. Check a sample of 10 to 20 leads from each version and record how many turned into a conversation or a job.

How to tell luck from a real gain

Any two pages will show different numbers by chance. The question is whether the gap is bigger than chance would usually produce. You do not need heavy statistics, but you need three rules.

  • Set the sample first. Evan Miller, a respected writer on testing, shows in "How Not To Run an A/B Test" that stopping the moment the result looks significant gave a false winner 26.1 percent of the time in his example, against the 5 percent expected. Decide how many visitors you need, then wait.
  • Know what a small test can show. These figures are derived by me, not from a vendor. I used the standard two-sided formula for comparing two rates, with equal groups, 95 percent confidence and 80 percent power: moving from a 5 percent rate to 7.5 percent needs about 1,500 visitors per version, but moving from 5 to 6 percent needs about 8,200. Check your case in a calculator such as Evan Miller's.
  • Use a significance calculator for the final number. Many testing tools show one. Remember that tools differ: Optimizely usually runs at 90 percent, while Convert uses 95 percent as its reference. Say which level you used.

If the gap is too small to call, write "not enough data to tell" and run it longer. That is a legitimate result.

A worked example with made-up numbers

These numbers are invented to show the arithmetic, not results from a real business.

An agency runs ads for a solar installer at $4,000 a month. The quote-request page converts at 3.0 percent. The agency changes only the headline and the first photo (a real crew photo instead of a stock panel image). The client agreed the goal (a submitted quote form), the sample and the rule in writing. To detect a rise from 3.0 to 5.0 percent, the formula above (95 percent confidence, 80 percent power) gives about 1,500 visitors per version, so that was the sample.

Version A (old)Version B (new)
Visitors (4 weeks, split 50/50)1,5201,510
Quote forms sent4676
Rate3.0 percent5.0 percent
Quotes that reached a site visit (sample of 20 each)8 of 208 of 20

The report says: "Version B produced 2.0 points more quote requests (5.0 against 3.0 percent) over four weeks and about 1,500 visitors per version. The lead quality in our sample was the same, so the extra requests look real. The sample is at the size we agreed. Nothing else changed in ads or budget. Recommendation: keep B and test the form length next." The agency does not say "B converts 50 percent better, guaranteed". It says what was measured, and what remains uncertain. If the same rates had shown up with only 300 visitors per version, the report would say "promising, not enough data".

Wording that is honest and still useful to a client

  • Say "in this test" and "over these weeks". Do not say "always" or "will".
  • Give both numbers and the sample, not only the percentage lift. "From 3.0 to 5.0 percent" is clearer than "up 67 percent".
  • Say what you cannot know. A short line beats a long disclaimer.
  • Never promise future results, and do not turn a test result into a forecast of leads or revenue.
  • If B lost, say it plainly and say what you learned. A tested loss is useful.
  • For benchmarks, name the source, year and sample. Unbounce's 2024 report gave a median landing page conversion rate of 6.6 percent across industries (41,000-plus landing pages), but that is a median across its own users' pages, and LocaliQ's 2026 search benchmarks show 8.18 percent for Google Ads search, from its own customers' campaigns. Do not use these to say a client's page "beat the average".

When there is not enough traffic for a split

Many small clients get a few hundred clicks a month. A fair split test is not possible, so use level 3 and be open about it.

  1. Record the 4 weeks before the change: clicks, enquiries, rate, cost per enquiry.
  2. Change one thing on the page. Keep the ads, budget and offer the same.
  3. Record the 4 weeks after, and list anything that differed, such as holidays or rivals' offers.
  4. Present both periods side by side, and say "consistent with the page helping, not proven".
  5. Repeat for a second period if you can. Two matching before and after results are stronger than one.

Another option is to make a bolder change, because a big gain is easier to see in a small sample than a tiny one.

Where the numbers can come from

SourceUse it forCost (as of November 2026)
Google Analytics 4 and Search ConsoleVisits, key events (actions you mark as important), and a before and after view. Google Analytics 4 is Google's free web analytics tool; Search Console is Google's free report on how your pages appear in searchFree
By hand: a sheet plus your form and call logsCounting enquiries by version and checking lead qualityFree, your time
Instapage and HubSpotInstapage, a landing page builder for ad teams, lists server-side A/B testing and traffic splitting on Optimize. HubSpot is a marketing and sales platform; check its pricing page for which plan includes content testing.Instapage: Create $79 a month billed annually or $99 monthly; Optimize $159 annually or $199 monthly; Convert custom. HubSpot: per seat, Professional $90 a month billed annually, $100 monthly
UnbounceA built-in split test with reporting; the Experiment plan lists confidence intervals and conversion reportingStarter $29, Build $99, Experiment $149, Optimize $249 a month billed monthly; Agency from $349 a month billed annually
PloyA section test, with an exposure event logged for each visitor assigned to a versionFree $0, Starter $50 a month, Pro $300 a month, Enterprise by quote

Google Optimize, once the free option, stopped on 30 September 2023. For other tools, best landing page builders for ads compares what each lists.

Where Ploy fits (one option of several)

With Ploy, you can run an A/B test on one section of a live page, with a traffic split from 10 to 90 percent and a goal. You write a hypothesis, choose the goal and a guardrail, and Ploy's Agent drafts the challenger. Ploy logs a built-in event named experiment.exposed when a visitor sees a version of a test.

Know the limits before you promise a client anything:

  • Experiments are switched on per workspace on request (email sales@ploy.ai).
  • One section of one page per test. The docs do not describe full-page splits.
  • The number of experiments at once is capped by plan, and the number is not published.
  • The statistics method and minimum sample are not published. Apply your own sample rule and use your own calculator for the final number.
  • Visitors who refuse cookies always see the current version and are not counted. Your results therefore describe visitors who accept cookies, which may differ from everyone. Say that in the limits paragraph.
  • Goal events must already be tracked and Ploy analytics must be on.

Ploy's pricing page shows Free $0, Starter $50 a month, Pro $300 a month and Enterprise by quote, as of November 2026. Ploy's docs say unused credits do not carry over. Check Ploy for agencies for what its agency offer includes.

Who should skip Ploy for this

  • You need a full-page test. Unbounce or Instapage are built for testing page variants. See Ploy vs Unbounce.
  • The client needs figures for everyone. If excluding cookie refusers would skew the audit, count with your booking system or form log instead.
  • You need a published statistics method. Ploy's docs do not describe one. A tool that shows confidence intervals, or your own calculator, is better for a report that may be challenged.
  • You have little traffic. No tool solves that. Use the before and after route and say so.

Copy-ready proof checklist

  1. Goal, change, sample and decision rule agreed in writing before launch.
  2. Both versions ran at the same time (or, if not, the report says it was a before and after).
  3. Same goal counted the same way on both versions.
  4. Same ads, budget and offer for both, or a list of what differed.
  5. Sample reached and at least one to two full business cycles.
  6. Lead quality checked on a sample from each version.
  7. Numbers shown with visitors, enquiries and rate, not only the lift.
  8. Limits paragraph written, including who was not counted.
  9. Next step and date included.

How I'd approach this

I would treat the evidence pack as part of the job, not as a sales tool. Write the goal, sample and decision rule down with the client before the change, use a split wherever traffic allows it, and show a clear "promising, not conclusive" when the sample is small. Clients usually forgive a modest result. They do not forgive a big number that falls apart a month later.

First-week plan: day 1, email the client the four agreement lines. Day 2, record the baseline for the last 30 days. Day 3, build the one change and check tracking on a phone. Day 4, launch at an even split or note the start of the before and after period. Days 5 to 7, check only for breakage. Decision rule: use a split tool (such as Unbounce, or Ploy's section tests if experiments are enabled on your account) when the page gets a few thousand clicks a month and you will test often; use the before and after route when it gets less than that.

If you want a result report that also records which booking or sale came from which version, Blur Studio builds that custom version, connecting tests to your booking or order system. See what that includes.

Related guide: testing a landing page headline.

Questions people ask

How do I prove to a client that a landing page change worked?

Run the old and new page at the same time, count one agreed goal the same way on both, and wait for the sample you set in advance. Then show visitors, enquiries and rate for each, what else changed, and the limits. If you cannot split traffic, show a before and after and say it is weaker evidence.

Is a before and after comparison enough?

It is useful but weaker, because ads, season, price and competitors all change over time. Use it when traffic is too low for a split, keep everything else the same, list anything that changed, and say the result is consistent with the page helping, not proven. A fuller method is in try a new version safely.

How many visitors do I need before I can report a result?

It depends on your starting rate and the lift you want to detect. By my own arithmetic at 95 percent confidence and 80 percent power, moving from 5 to 7.5 percent needs about 1,500 visitors per version and from 5 to 6 percent needs about 8,200. Use a calculator such as Evan Miller's for your numbers.

What if the new page lost?

Report it plainly. A tested loss is useful: you kept the better page and learned what not to change. Show the same evidence pack, say what you will try next, and keep the old page live.

Does Ploy report A/B test results to clients?

Ploy's experiments log an exposure event (experiment.exposed), and the docs describe a compare view for the two versions before launch and a warning when the challenger is likely doing worse. Its docs do not publish the statistics method or minimum sample, and experiments are on request, so build your own sample rule into the report. See Ploy for current details.

Can I use an industry average to show the page beat the market?

Be careful. Benchmarks come from one vendor's customers, use different definitions of a conversion, and are a median or a mean depending on the source. Unbounce's 2024 median across industries was 6.6 percent and LocaliQ's 2026 figure was 8.18 percent. Use them as a rough check, never as proof.

Do visitors who refuse cookies affect my test results?

Yes for some tools. In Ploy, visitors who refuse cookies always see the current version and are not counted, so results describe visitors who accept cookies. Your own form and call logs still capture everyone, so compare the two. Check local privacy law and read is website visitor tracking allowed under privacy law.

More in this topic: Websites: testing and proving page changes (4 guides) →

Sources

Ready for the next step?

A 30 minute call. If you do not need me, I will tell you.

Ranjit Bhinge You talk to me, Ranjit. I reply within a few hours.

Not sure what you need? Tell me what is going on. I will point you the right way, even if it is not me.

Blur Studio · 30 minute call

Book a 30 minute call

You talk to me, Ranjit. If you do not need me, I will tell you.

  • We talk about your business and what is going on.
  • You leave with a clear next step, even if it is not me.
  • I reply to every message within a few hours.
Loading the calendar…