The attribution report in your Google Ads account was built to make Google Ads look good, and it does that job well.
The conversions in it are real. The lie is the same one a photograph tells: one angle, with everything that happened before the frame left out.
A typical enquiry to a UK firm starts with a generic search, then a few days of nothing, then a LinkedIn post or a recommendation, then a search for the company by name. The last click report hands all of the credit to that final brand search. The person who ran the generic search campaign gets nothing.
This article covers what Google changed in 2023, why Google Ads and GA4 disagree about the same month, and a three-part model that a small firm can run without a data scientist.
What Google removed in 2023, and what is left
Google Ads now offers two attribution models, last click and data-driven, because it retired the other four in 2023. Google announced on 6 April 2023 that first click, linear, time decay and position-based attribution were going away in both Google Ads and GA4.
The timeline ran like this:
- May 2023: the four models stopped being available for new GA4 conversion actions.
- June 2023, pushed to mid-July: they stopped being selectable for new Google Ads conversion actions.
- September 2023: any Google Ads conversion action still using one of them was moved to data-driven attribution, and the models were removed from the Overview page and the Model comparison report.
- November 2023: GA4 followed, and its help page now states that the four models "are no longer available as of November 2023".
Google's reason, in its own words, was that "these models don't provide the flexibility needed to adapt to evolving consumer journeys". It also said that under 3% of Google Ads web conversions used the retiring models in February and March 2023, so for most accounts nothing visibly changed.
That is the problem. Most UK SME accounts were on last click before 2023 and are on last click or data-driven now, and nobody has ever asked what either number means. Google's current help page says data-driven is the default for most conversion actions, and that the model you pick changes the Conversions column and therefore every bid strategy that reads from it.
Why Google Ads and GA4 disagree about the same month
Google Ads and GA4 report different conversion counts for the same month because they count on different dates, over different windows, with different channels allowed to take credit. Google itself says that discrepancies "often up to 20%" are expected between the two.
The date problem catches most people first. Google Ads conversion columns are "calculated based on the time of the click, not the time of the conversion". A click on 5 May that turns into an enquiry on 10 May lands in Google Ads on 5 May. GA4 puts the same enquiry on 10 May. Run a report for the week of 6 to 12 May and the two tools will never agree.
The windows differ too. A Google Ads click-through conversion window defaults to 30 days, with engaged-view at 3 days and view-through at 1 day. GA4's lookback for most key events defaults to 90 days, with 30 days for acquisition events. A 45-day sales cycle shows up in one tool and not the other.
Then there is the question of who is allowed to take credit. Google Ads defaults to "Google paid channels", so only its own ads compete for the conversion. GA4 defaults to "paid and organic channels", so an organic click or an email can win. Same enquiry, two different winners, and both reports are technically correct.
Google's own advice for lining the two up is to set the Google Ads columns to "All conv." or "All conv. (by conv. time)" and check both accounts are on the same time zone. Even then, more than 95% of GA4 key events are attributed within 14 days, so a comparison on fresh data is a comparison on incomplete data.
Here is how the main views compare for a firm whose enquiry started with a generic search ad and ended with a brand search.
| Attribution view | Who gets the enquiry | Counted on | Window | What it is good for |
|---|---|---|---|---|
| Google Ads, last click | The last Google ad clicked (usually the brand keyword) | Date of the click | 30 days by default | Spotting which ad closed the deal |
| Google Ads, data-driven | Split across Google ads by Google's model | Date of the click | 30 days by default | Feeding automated bidding |
| GA4, last click (paid and organic) | The last channel of any kind, including organic and email | Date of the enquiry | 90 days by default | A whole-site view of the closing channel |
| GA4, data-driven (paid and organic) | Split across all channels by Google's model | Date of the enquiry | 90 days by default | Whole-site view, with fractional conversions |
| "How did you hear about us" field | Whatever the customer remembers | Date of the enquiry | Unlimited, but memory fades | Finding the channels no tracker can see |
| Branded search volume | Nobody, by design | Monthly | Rolling | Telling whether the total demand is rising or falling |
None of these is wrong. Each one answers a different question, and the mistake is treating any single row as the whole picture.
What a real path looks like
A real enquiry path has more touches than any of the reports admit, and most of them are invisible to the trackers. The flow below is a typical one for a UK professional services firm. It is a worked example, not a single client account.
flowchart LR
A[Generic search: 'employment solicitor Northampton'] -->|clicks Search ad| B[Reads a service page, leaves]
B -->|4 days| C[Sees a partner's LinkedIn post]
C -->|no click| D[Asks a colleague, gets the firm's name]
D -->|day 9| E[Searches the firm's name]
E -->|clicks organic listing| F[Submits enquiry form]
F --> G{Who gets credit?}
G -->|Google Ads last click, paid only| H[Brand ad, if one ran. Otherwise nothing]
G -->|GA4 last click, paid and organic| I[Organic search]
G -->|'How did you hear' field| J[A colleague recommended you]
Google Ads sees the first click and, if a brand campaign is running, the last one. It never sees the LinkedIn post or the colleague. GA4 sees two sessions and gives the enquiry to organic search. The customer, if asked, says "a colleague".
Data-driven attribution does not fix this. It can only split credit between the touches it can see, which in Google Ads means Google ads. The LinkedIn post and the recommendation are not in the dataset, so they get zero credit under every model Google offers.
What data-driven attribution can and cannot tell you
Data-driven attribution tells you how Google's model weighs your own Google ads against each other, and nothing else. Google describes it as using "your account's data to calculate the actual contribution of each interaction across the conversion path", and says advertisers who switch typically see a 6% increase in conversions.
Read that 6% carefully. It is a rise in reported conversions, because the model counts fractions of conversions that last click would have given wholly to the brand term. The number of enquiries in your inbox does not change when you change the model.
The formula is also private. As Search Engine Land noted at the time of the sunset, the data-driven formula differs by advertiser and is not visible, so you cannot check its working. For an account bidding on target CPA, that is acceptable: the machine bids on its own numbers. For a managing director deciding whether to keep paying for generic search, it is not an answer.
We use data-driven attribution in every account we bid automatically on, and we do not use it to judge channels. Those are two different jobs, and the second one needs data from outside Google Ads.
A simpler model for a small firm
The model we recommend has three parts and needs nothing beyond a spreadsheet. The parts are a "how did you hear about us" field on the enquiry form, branded search volume as the aggregate signal, and a monthly channel-by-channel read that puts the numbers side by side.
Part one: ask the customer
Add one question to every enquiry form, phone script and discovery call: how did you hear about us? Make it a free text field, not a dropdown, because a dropdown only lets people pick channels you already know about.
The evidence for this is better than you might expect. HockeyStack analysed 8,528 self-reported attribution responses from demo forms at B2B companies and compared each answer with the last-touch channel its software had recorded. Where software said "search", the customer agreed 53% of the time. Where software said "social", 60%. Where software said "email", 21%. Where software said "display", 2.5%.
Word of mouth was the third most common answer at 18% of responses. No tracker reports word of mouth. The report's own framing is the right one:
"Think of self-reported attribution as the most remembered touchpoint of the person who books the demo." HockeyStack, Self-Reported Attribution Report
The same report warns that "Google" as an answer could mean an ad, a brand search or a blog post, and that about 20% of responses were unusable. So the field is a signal, not a ledger. Treat it the same way you treat a reference from a previous employer: useful, directional, not proof.
One further example from outside Google Ads. Graphite compared GA4 last-touch with a post-purchase survey for the software company n8n and found GA4 gave Google search 7% of conversions while customers gave it 30%. That is a single company, reported by an agency with a product to sell, so weigh it lightly. But the direction matches HockeyStack's.
Part two: watch branded search
Branded search volume is the one number that every channel feeds into and no channel can claim. If the LinkedIn posts, the radio ad, the generic search campaign and the referrals are working, more people type your firm's name into Google. If they are not, the line is flat.
Pull it monthly from Google Search Console (impressions for queries containing your brand) and, if you run a brand campaign, from the Google Ads search terms report. Plot it against your spend by channel on the same chart. A channel that costs money and never moves the brand line is a channel to question.
This is also the fairest test of generic search, which last click always undercounts. If you pause generic search for a month and brand searches drop, the generic campaign was doing its job regardless of what the attribution report said.
Part three: the monthly honest read
The monthly read is one table with a row per channel and four columns. The columns are spend, Google Ads or GA4 conversions, enquiries that named the channel in the "how did you hear" field, and the firm's judgement in one sentence. The judgement column is the point. It forces someone to write "brand search is harvesting demand the LinkedIn posts created" instead of reading a CPA off a screen.
A few rules keep it honest. Compare months only when the data is at least 14 days old, because of the attribution lag. Use the "by conv. time" columns in Google Ads so the dates line up with GA4 and the form entries. And count enquiries, not conversions: a conversion is a form fire, an enquiry is a person you could call back.
When the three sources disagree, and they will, the disagreement is the finding. If Google Ads credits the brand campaign with 40 conversions and only four customers said "Google ad", the brand campaign is collecting demand, not creating it, the same way Performance Max can quietly buy your brand searches and report them as wins. That is worth knowing before the next budget conversation.
The decision rule
Use the model Google gives you to run the bidding, and use the model the customer gives you to run the business. Any report that can only see Google's own clicks will flatter Google's own clicks, and no amount of switching between last click and data-driven changes what it cannot see. When you read a UK Google Ads agency's monthly report, ask how much of the credited performance is really branded search.
If you want someone to set up the form field, the branded search tracking and the monthly read, and then run the paid search behind it, our paid media team does this as standard, or contact us to talk it through.