On this page
Before you change anything
Write down what “stopped working” means numerically, because the answer changes the whole investigation. Impressions down is a completely different problem from impressions steady and clicks down, which is different again from clicks steady and enquiries down. Pull a ninety-day daily chart of impressions, clicks, cost and conversions, and find the date the line changed.
Then check the change history before anything else. In the Google Ads interface it sits under Tools, and it lists every edit with a date and a user — including edits made automatically by Google. A surprising share of mystery collapses resolve in ninety seconds because somebody, or something, made a change on the exact day the line moved. If your ad account was inherited, this is also where you discover what the previous manager did on their way out.
One discipline throughout: change one thing, then wait. Trade accounts are low-volume enough that two weeks is the shortest honest read on most changes, and if you alter five settings on Monday you have destroyed your own evidence.
Causes one to three: the market moved
1. Your season turned and you did not notice. Nearly every trade has one, and it rarely aligns with the calendar quarters. Compare this month against the same month last year rather than against last month. Google Trends will show you the shape of demand for your main search term over five years for free, though be aware it reports relative interest rather than absolute volume, and in a small town the line is noisy enough to mislead. If last September looked exactly like this September, the account is fine and your expectations were wrong. We have written up how to plan for this on the seasonality guide.
2. Somebody new entered your auction. Open the auction insights report, segment it by month, and look at the impression share and overlap rate columns over the last six months. A new advertiser bidding hard, or a national franchise switching its geography on, will pull your impression share down without a single thing changing in your account. This is not always worth fighting. If a well-funded competitor is buying the most contested head term at a loss, the better move is to concede that term, take the longer and more specific searches they have ignored, and let them pay for the traffic that converts worst.
3. The search itself moved. Query behaviour shifts. People start describing the same job differently, or an AI overview begins answering the informational half of the query so the click never happens. Check your Search Console impressions over sixteen months alongside your ads data: if organic impressions for the same phrases fell in step with your paid impressions, demand moved rather than your account breaking.
There is a fourth external possibility that is not on the numbered list because it is not really an advertising problem, and it is worth ruling out early. Something changed about the business that the market noticed: a price increase, a run of poor reviews, a rebrand, a phone number that now goes to a different office. Ads bring people to a decision; they do not make the decision. If the click volume held and the enquiry rate fell across every campaign at once, including brand searches from people who already knew you, the cause is more likely to be what happens after the click than anything in the account.
Causes four to six: the account changed under you
4. Auto-applied recommendations. This is the single most common cause we find in inherited accounts. Google can automatically apply its own recommendations unless you switch them off, and the two that do real damage are “use broad match keywords” and “add new keywords”. Overnight, a tight account bidding on “emergency electrician Solihull” starts matching “electrician apprenticeship” and “how to wire a socket”. Go to the recommendations page, open the auto-apply tab, and turn every toggle off. Then read the search terms report for the period since it happened and add the damage as negatives.
5. Conversion tracking broke. Enquiries may not have fallen at all — you may just have stopped counting them. The usual culprits are a website rebuild that dropped the tag, a consent banner change that now blocks the tag until someone accepts cookies, a call tracking subscription that lapsed, or a form plugin update that changed the thank-you URL. The tell is that conversions fall off a cliff on a single day while clicks and impressions carry on as normal. Real demand changes are gradual; tracking failures are vertical. Use Google Tag Assistant to load your own site and confirm the conversion actually fires, and cross-check against the enquiries in your inbox for the same week.
The second version of this is subtler and worse: tracking still works, but it is counting the wrong thing. If somebody set a contact page view as a conversion, the account has spent months optimising towards people who look at your phone number and leave. Our note on setting up call and lead tracking covers what should and should not be counted.
6. The bidding strategy is strangling delivery. Target CPA and Target ROAS are throttles as much as they are optimisers. If somebody set a target below what the auction will actually deliver — often after one good week — the strategy responds by simply not entering the auction, and you see impressions collapse while the reported cost per conversion looks beautiful. Check the strategy's recommended target against the one that is set. The same applies to a portfolio bid strategy quietly spanning campaigns with wildly different economics.
Causes seven to nine: the enquiry arrives and dies
7. The landing page changed or slowed down. Website rebuilds are a common hidden cause. A new page that looks better may have moved the phone number below the fold, replaced a simple form with a multi-step one, or added enough script to push mobile load time past three seconds. Compare conversion rate by device before and after the rebuild date. If mobile fell and desktop did not, it is nearly always speed or layout. What a trade page has to get right is set out in the landing page guide, and if the site itself is the constraint that is a web design problem rather than an advertising one.
8. Nobody is answering. Blunt, and more common than any technical fault. If calls generated at 4pm on a Friday go to voicemail, or the office mobile is in a van with no signal for six hours, the ads are working perfectly and the money is still gone. Pull the call log against the ad schedule. Every unanswered call during a period you are paying for is a decision to buy something and throw it away. Automatic missed-call text-back is the cheapest fix in marketing and takes an afternoon to set up, which is part of what CRM automation is for.
9. Lead quality drifted while the numbers held. The account reports the same forty conversions a month, and none of them turn into work. This is what happens when broad match, loosened geography or a conversion action that fires too easily gradually shifts the mix towards cheaper, worse enquiries. The only way to see it is to reconcile: take last month's conversions, mark each one as quoted, not quoted or nonsense, and compare against the same exercise three months earlier. If the proportion of nonsense has doubled, the account is optimising towards it.
Symptom to cause, at a glance
| What you can see | Most likely causes | First place to look |
|---|---|---|
| Impressions fell sharply on one date | Bid strategy target, budget change, disapproved ads, payment failure | Change history, then the campaign status column |
| Impressions fell gradually over weeks | Seasonality, a new competitor, declining query volume | Auction insights segmented by month |
| Impressions up, click-through rate down | Broad match pulling in irrelevant queries | Search terms report for the period |
| Clicks steady, conversions to zero overnight | Tracking broken — tag, consent banner, thank-you URL | Tag Assistant, and your actual inbox |
| Clicks steady, conversions falling slowly | Landing page change, site speed, form friction | Conversion rate split by device |
| Conversions steady, no jobs | Quality drift, wrong conversion action, unanswered calls | Reconcile conversions against quotes sent |
| Cost per conversion climbing steadily | Auction pressure, or you have exhausted the good searches | Impression share on core terms |
The order to work in
- Confirm the ads are actually running. Billing, campaign status, ad approval status, and whether an ad group has zero eligible ads.
- Read the change history for the fortnight before the line moved.
- Verify the conversion tracking fires, by submitting your own form and ringing your own number.
- Read the search terms report for the affected period and add negatives for everything that appeared.
- Check the auto-apply recommendation toggles and switch them off.
- Compare auction insights month on month.
- Compare conversion rate by device against the date of any website change.
- Pull the call log and count the unanswered calls in paid hours.
- Reconcile a month of conversions against quotes actually sent.
Nine steps, most of them ten minutes each. It is genuinely rare to reach step nine without having found the problem.
When the answer is to start again
Occasionally the honest conclusion is that the account is not repairable in place. The signs are consistent: a structure nobody can explain, hundreds of keywords in a handful of ad groups, a conversion history built on the wrong conversion action, and several bid strategies that have been chasing bad data for a year. In that situation, incremental fixes fight against a machine learning model that has been trained on the wrong outcomes.
Rebuilding is not free. You lose the conversion history and pay a re-learning period of a few weeks. It is worth it when the existing history is actively misleading, and not worth it when the structure is sound and the problem is one broken setting. We would rather fix than rebuild, and what we have measured in scaffolding, roofing, glazing and surfacing accounts has more often come from removing waste than from starting over. If you want a second opinion before deciding, ask for one — and if the account is basically fine and the market simply turned, we will say that too.
Rather have someone else do this?We run exactly this for trade and construction firms across the UK. Call and we will tell you what is realistic in your area, including when the answer is that it is not worth it.
Call 07443 392243WhatsAppCommon questions
My impressions dropped to almost nothing overnight. Where do I look first?
Billing and ad approval, in that order, then the change history. A declined card, a policy disapproval across every ad in a group, or a bid target somebody lowered will all produce a vertical drop. Genuine demand changes are gradual, so a cliff edge almost always means a setting or an account status rather than the market.
Could a Google algorithm update have caused it?
Core search updates affect organic rankings, not the paid auction directly. What can change paid performance without any action from you is a shift in how results pages are laid out, a new competitor entering, or automated features being switched on in your account. Check auction insights and the auto-apply settings before blaming an update.
Is it worth pausing everything while I investigate?
Usually not. Pausing loses the recent conversion data that automated bidding depends on and adds a re-learning period once you restart. The exception is when you have found genuine waste you cannot stop any other way, such as a broad match campaign burning budget on job-seekers while you rebuild the negatives.
How much variation is normal week to week?
More than most owners expect. A local trade account producing thirty enquiries a month can easily swing between four and twelve in a given week on nothing but weather and school holidays. Two quiet weeks is noise. Six weeks of decline against the same period last year is a trend worth investigating.