Paid traffic dies on the page, not in the auction

When a Google Ads account underperforms, almost everyone audits it from the wrong end. Bids, match types, Quality Score, the bidding strategy, whether someone should be testing Broad. All of that is upstream of the actual failure, which usually happened three seconds after the click, on a page that opened with a slogan rather than an answer and then asked a complete stranger for eleven fields of personal information.

Here is the test we start with, and you can run it in two minutes. Take your highest-spending keyword, search it on your phone, on mobile data rather than office wifi, and click your own ad. Then count the seconds until you can tell what is being offered and what you are supposed to do next. Most owners have never done this. They have seen their landing page on a desktop, logged in, with the images cached.

Paid visitors are a worse audience than organic ones

This is not flattering but it is true. Someone arriving from an organic search result chose you from a list and has some intention to read. Someone arriving from an ad was interrupted, is probably comparing you with two other tabs, and has no prior knowledge of your company at all. They arrive with the intention of leaving unless given a reason to stay.

So the first screen has to do the whole job. In the accounts we have audited, these are what reliably kill paid traffic:

Message mismatch. The ad said emergency roof repair. The page says welcome to our family business, established 1998. The visitor cannot see their own search reflected back and leaves.
Load time on a real phone. Google’s own threshold for Largest Contentful Paint is 2.5 seconds at the 75th percentile, with Interaction to Next Paint under 200 milliseconds and Cumulative Layout Shift under 0.1. INP replaced First Input Delay as a Core Web Vital in 2024, which matters because a page can load quickly and still feel broken when you tap something.
Total silence on price. You do not have to publish a price list. You do have to indicate a range, a starting point, or how the pricing works. Silence reads as expensive and as evasive at the same time.
A form built for your CRM rather than for the visitor. Every field is a separate decision to keep going. Name, a way to reach them, and one line about the problem is enough to start a conversation.
No evidence. No photographs of real work, no named reviewer, no address. Paid traffic is strangers by definition, and strangers want proof.
The full site menu, sitting at the top of the page as an escape route.
A phone number that is an image, or three clicks deep, in a trade where people want to call someone now.

A note on the statistic everyone quotes

You have probably been shown the figure that 53 percent of mobile visits are abandoned when a page takes more than three seconds. It comes from a Google and SOASTA analysis published in 2017, built on aggregate data from over 10,000 mobile domains. It is correlation across a very large sample, not a controlled test, and the two companies behind it sold advertising and performance monitoring respectively. Slow pages also tend to be heavy, ad-stuffed, badly built pages, and that confound was never separated out.

We do not use the number and we would rather you did not either. Page speed matters for reasons you can verify in your own analytics in an afternoon. It does not need a nine-year-old vendor statistic propping it up.

The arithmetic

Now the part that makes this the cheapest work in the account. Put your own figures in.

Clicks at 4 lev, a page converting at 2 percent: every lead costs you 200 lev. The same page at 4 percent: 100 lev. Nothing in the auction changed. No new budget, no bid strategy, no negotiation with Google. You simply stopped losing half of the people who had already agreed to come and look at you.

Conversion rate is the only lever in paid search where the improvement compounds against every click you will ever buy again, and it is the one most owners never touch, because it lives on the website and the website feels like it belongs to whoever built it.

Supermarket self-checkouts are a good example of the same mistake in a different trade. Every design decision goes into the scanning: the beep, the weight sensor, the screen prompts, the layout of the bagging area. Almost none of it goes into the thirty seconds after something will not scan, which is the only part of the interaction anybody remembers.

The order we fix things in

Read the search term, then read your headline. Does the headline answer that exact query in plain words? Rewriting it is usually the highest-return change available and takes ten minutes.
Cut the form to the minimum needed to make contact. Qualify in the conversation.
One clear action above the fold, repeated further down. One, not four competing ones.
Move the proof next to the action instead of leaving it on a separate About page.
Build a page per major service. Sending every keyword to the homepage puts a permanent ceiling on your conversion rate, because a homepage is the wrong instrument for a specific intent.
Give each change enough traffic to mean anything. A page that converted better across fifty visits has told you nothing at all.

When it is not the page

Sometimes the traffic really is the problem. If your search terms report is full of research queries, students and job seekers, no page will rescue it and a week spent rebuilding one is a week wasted. Check the query mix before you touch the design.

Sometimes it is the offer, and a better page only makes that visible faster. That is useful information rather than a conversion fix.

And sometimes the page works and the follow-up does not. Leads sitting unanswered for two days convert far worse than leads answered in minutes, which nobody disputes. Be careful with the numbers attached to it, though: the famous speed-to-lead multipliers come from a 2007 study run by InsideSales with MIT, and a 2012 Velocify report, and they are routinely miscredited to Harvard Business Review, which published a separate piece on the subject in March 2011. Three different studies from three different years, welded into one statistic. Treat the direction as sound and the figures as marketing.

We have seen accounts blamed for months when the real failure was an inbox nobody was watching. That one is not a website problem either.