How long does Google Ads take to work?
by Harry Watts · · updated · 7 min read
Three different questions hide inside this one. When does traffic start? Hours. When does the data mean anything? Weeks. When can you fairly judge the account? 60–90 days. People who conflate them either quit too early, or get strung along for months.
The sellers answer "immediately". The agencies protecting a retainer answer "six months". Both are self-serving. Here's the stage-by-stage version, with what to look at during each wait so none of it is blind.
Hours: traffic, and the checks that can't wait
Ads typically serve within hours of approval. Day one isn't about performance; it's about confirming the machinery:
- Impressions flowing: if not, check ad approval status and budget, not the copy.
- A test enquiry through your own site, recorded exactly once: this should predate launch, but day one is the last cheap moment to catch it.
- The first search terms: a glance, not a purge. You're checking the targeting caught roughly the right river.
Weeks 1–4: learning, and what "learning" actually means
Smart bidding spends its early weeks calibrating: estimating, auction by auction, how likely each click is to convert for you. The interface shows "Learning" for about a week after a strategy starts, but the practical version runs until the account generates roughly 30 conversions a month. Until then, expect volatile CPCs and uneven days. That's calibration, not failure.
What to do during it, and, more importantly, what not to:
- Do the weekly search-term half hour: maintenance never resets learning, and early exclusions compound.
- Don't re-engineer. New bid strategies, big budget swings and rebuilds each restart the calibration you're waiting for. Panic-editing in week two is the single most damaging habit in new accounts: every tweak resets the clock you're complaining about.
- Log, don't react. Note what you'd change; batch it for the month-two review.
Clicks but no conversions yet? At a 5% conversion rate, 60 clicks converting zero times is ordinary chance. Verify the tracking, check the search terms are relevant, test the page on a phone, and if all three pass, the honest reading is "not enough data", not "doesn't work".
Weeks 4–8: patterns, or a diagnosis
By now, one of two things is true. Either conversions are arriving and a cost per acquisition is taking shape, or volume is far short of the ~30/month run-rate, and the wait itself won't fix that. Short volume has three usual causes, checked in order:
- Budget too thin for the market: buying too few clicks to convert at any realistic rate. The arithmetic says whether more budget or narrower targeting is the answer.
- Clicks that don't convert: a page or offer problem wearing a campaign costume.
- Conversions happening but not recorded: still the most common of the three.
Day 60–90: the verdict you pre-agreed
This is where the number you wrote down before launch earns its keep. Compare cost per customer against it: level and direction. Three honest outcomes:
- Clears the target: scale from proven headroom, and start the second-platform conversation.
- Near and improving: continue with specific fixes. This is the commonest 90-day position and a good one.
- Nowhere close, flat trend: stop or restructure. The account has answered; respect the answer.
"Still learning" — legitimate, or an excuse
Legitimate: for about a week after structural changes (new bid strategy, conversion-action changes, major budget moves). Every reset restarts it, which is exactly why changes should be deliberate and batched.
An excuse: as the standing explanation for a flat account across months. Learning is a phase with an exit, not a climate. If someone has said "it's still learning" in three consecutive reviews, open the change history: it will show either the run of resets that justify the phrase, or the silence that doesn't.
A caution about fast results — in both directions
Week-two triumphs are usually brand traffic and low-hanging fruit dressed as strategy; week-two disasters are usually noise. Small numbers lie in both directions, which is why the sellers' "immediately" and the quitters' "it didn't work" are so often the same statistical mistake with different signs. The 60–90 day window isn't patience for its own sake — it's the minimum sample at which the account is telling you the truth.
Quick answers
How long until it works?
Traffic in hours. Meaningful data in two to four weeks. A fair verdict at 60–90 days, once smart bidding has ~30 conversions a month to work with.
What's the learning period?
Smart bidding calibrating its conversion predictions on your data. "Learning" shows for about a week after changes; practically it runs until ~30 conversions a month. Volatility during it is calibration, not failure.
Clicks but no conversions: broken?
In the first fortnight, usually just small numbers. Verify tracking with a test enquiry, check search-term relevance, test the page on a phone. All pass? Not enough data yet.
When can I make changes?
Maintenance (search terms, negatives) weekly from day one: it never resets learning. Structural changes reset it, so batch them. Panic-editing restarts the clock you're waiting on.
Still learning, or just failing?
Stage gates: week two = machinery working; weeks 4–6 = conversions trending to ~30/month; day 60–90 = CPA vs your pre-agreed number. "Learning" for months with a flat change history is an excuse.
Somewhere in this timeline and unsure which stage you're at
Thirty minutes on a free intro call places you: machinery fault, data shortage, or verdict time. Each has a different fix, and two of the three don't need anything bought. If you're paying someone and hearing "still learning" on repeat, bring the change history: the audit reads it professionally.