The traffic chart is flat. The conversion chart is not. Fewer people are buying or signing up, and the same number are arriving.
There are two stories that fit, and they send you to different people. In the first, the visitors changed: a campaign started sending people who were never going to convert. In the second, the site changed: something now stops people who used to finish. One is a marketing fix and the other is an engineering fix.
You can tell them apart in about half an hour, without guessing. This guide gives you the order of checks.
First, make sure the drop is real#
A conversion rate has two numbers in it, and either one can be wrong.
Count the conversions somewhere else. Orders in your shop backend, signups in your database, payments in your payment provider. If those counts held and your analytics shows a drop, nothing happened to your customers. The measurement broke. A renamed event, a missing tag on a new page, or a consent change can each do it. Our guide to the June 2026 Google Ads consent change covers one common version.
If the backend count fell too, the drop is real. Keep going.
A steady total can hide a changed mix#
"Sessions are steady" is a statement about a sum. The parts can move in opposite directions and leave the sum where it was.
Say paid social doubled this week and organic search fell by the same number of visits. The total did not move. If search visitors convert several times better than social visitors, your overall rate fell, and nothing on the site is broken.
So the first cut is always the same: break the flat total into its sources and look at each one against the week before.
Five cuts that separate the two stories#
Run these in order. Each one takes a few minutes in any tool with funnels and segments. Use the same two periods for all of them, the bad week and the last normal one.
1. By source and campaign#
Compare the conversion rate of each source across the two periods.
- One source fell and the others held. The problem travels with that source. Look at what changed in the campaign: audience, placement, creative, landing page.
- Every source fell together. The visitors did not all get worse at once. The problem is on the site.
Also check the share of each source. A source with a normal rate and a much larger share will pull the total down by itself.
2. By funnel step#
Build the funnel with real steps: landing, product or pricing view, start of checkout or signup, payment or form submit, success.
- The loss is at the first step. People arrive and leave. That is the pattern of visitors who did not want what the page offers.
- The loss is at one late step. People wanted it enough to get that far, then stopped. Interested visitors do not all change their minds on the same screen. Something on that screen is in their way.
If you do not have the step counts in one place, enter them in the funnel drop-off calculator to see which step lost the most.
3. By device and browser#
Split the affected step by device, then by browser.
- The drop is spread evenly. This fits either story, so rely on the other cuts.
- The drop is on one device or one browser. This is almost always the site. A layout change, a payment script, or a form validation that fails in one browser. Audiences rarely get worse on Safari only.
4. By new and returning visitors#
Returning visitors already know you. If their rate fell as much as the rate for new visitors, traffic quality is a weak explanation, because their quality did not change.
5. By time#
Find the hour the rate changed, not the day.
- A sharp edge. Line it up against your release log, your tag manager history, and the status pages of your payment and form providers.
- A slow slide over days. This fits a campaign that scaled up, a ranking change, or a seasonal shift.
The two signatures side by side#
| What you see | Bad traffic | Broken checkout or form |
|---|---|---|
| Where the loss sits | First step | One late step |
| Sources affected | One source or campaign | All of them |
| Devices and browsers | Even | Often one |
| Returning visitors | Rate holds | Rate falls too |
| Shape over time | Follows a campaign change | Sharp edge at a release or outage |
| Who fixes it | Marketing | Engineering |
Most real cases match one column on four or five rows. If yours matches both, you may have two problems that started in the same week. Treat them as two investigations.
Get the cut that names the cause
Ask Neo why conversion fell. Neo compares the segments and the funnel steps, and attaches the session recordings as proof. 14 days free, no card.
When it points to traffic#
Confirm it before you change anything.
Open the campaign that grew and check where its visitors land. Check that the landing page loads on a phone. Check that the tags on the link survive the redirect, because a campaign that loses its tags is reported under another source and confuses every cut above. The UTM builder opens a link the way a phone does after an ad tap and shows which tags arrive.
Then judge the campaign on its own numbers: visits, conversions, cost. A campaign that brings visitors who do not convert is a cost. Narrow the audience or stop it. The site does not need a fix.
When it points to the site#
Now you need to see what happens on the broken step.
Watch ten recordings of sessions that reached the step and did not finish, filtered to the device or browser that fell. You are looking for the same thing happening more than once: a button that does not respond, an error that appears after submit, a field that rejects valid input, a payment option that never loads.
A heatmap of the step helps when the recordings are not conclusive. Clicks on something that is not clickable, or many clicks on one button, mark the spot.
Then find the change. Compare the start time with your releases. Include changes that are not code releases: a new tag in the tag manager, a consent banner update, a changed setting at the payment provider. Our walk-through of checkout abandonment analysis goes deeper on this part, and the full conversion drop checklist lists the other causes.
A mobile-only drop#
A sudden drop on mobile, with desktop unchanged, deserves its own note, because it is common and it is usually misread as "mobile traffic is low quality".
Mobile visitors did not get worse overnight. Check three things:
- Mobile browsers against each other. If one browser fell and the others held, you have a compatibility bug.
- In-app browsers. Links opened from social apps load in the app's own browser, which behaves differently from the phone's default browser. Payment pop-ups, redirects and saved logins fail there more often.
- The last layout change. A button that moved below the screen edge, a banner that covers the submit button, a keyboard that hides the field in focus. None of these raise an error.
If the drop began the day you changed analytics tools#
This is a separate case, and nothing may be wrong.
Two tools rarely agree, because they define the two numbers in the rate differently.
- What a session is. Tools end a session after different idle times and treat a new campaign click differently. More sessions for the same visits means a lower rate.
- Who is counted. A cookie-based tool does not see visitors who decline the banner. A cookieless tool does. If the new tool sees more visitors, and those extra visitors convert less often, the rate falls and the count of conversions does not.
- Bots. Each tool filters automated traffic by its own rules.
- What a conversion is. One tool may count every thank-you page view. Another may count one conversion per visitor.
To check, pick three days when both tools ran. Compare raw visits and raw conversions, not the rates. If the conversions match and the visits do not, your business is the same and your denominator changed. Set a new baseline in the new tool and compare it only with itself from then on.
How this looks in GrainQL#
Each cut above is a filter or a question in GrainQL.
Mission Control shows each goal, how it changed against the period before, and where the conversions come from. A track draws the paths between a start event and a goal, and you can filter it by device, browser, source, country or operating system. Session recordings and heatmaps show what happened on the step that lost people.
You can also ask Neo:
- "Conversion rate fell this week and sessions did not. What changed?"
- "Compare checkout completion on mobile Safari with other browsers, this week and last week."
- "Show me sessions that reached payment and did not finish."
GrainQL is cookieless by default, with visitor IDs that rotate every day. A visit does not depend on an accepted cookie, so visitors who decline a banner are still in the counts.
Find out which story is yours
One tag, no cookie banner needed for GrainQL, 14 days free, no card.