SEO tools: when to trust GA4 multi-touch attribution

What to take away

Browser limits and consent tools now block much of the cross-site data that older attribution reports relied on, so the same suite can show fewer conversions than before. Most SEO suites, including Ahrefs and Semrush, foreground rankings, backlinks and keyword data or leave attribution to GA4; that often credits the final branded search and hides earlier research.

Multi-touch needs a durable identifier, usually a gclid or a server-side Measurement Protocol event, before the numbers mean anything. Match the model to the decision: content budgets justify data-driven credit, while paid channel splits often survive on last click. Record the rule, lookback window, date and owner. For licence choices, see which SEO tools are worth buying.

How search attribution methods differ

Attribution models are rules for assigning conversion credit to touchpoints. GA4 now offers data-driven, paid and organic channels last click, and Google paid channels last click. The older single-touch and rule-based models are mostly retired, so the practical comparison is between data-driven and last click.

Single-touch models: first click and last click

First click gives the whole conversion to the earliest recorded touch. Last click gives it to the final one. Both are single-touch rules: each gives the whole conversion to one recorded interaction. First click is no longer selectable in GA4, but last click remains available as a reporting model.

Last click flatters brand search and retargeting. A user who reads three guides, compares prices, then searches your brand name, produces a report where the brand search earns everything and the guides earn nothing. A 30-day and a 90-day window credit differently, so state the window whenever you name the model. See SEO tools measurement mistakes.

Multi-touch models: common forms and GA4's current option

Linear splits credit evenly across every recorded touch. Positional weights the first and last interactions more heavily and shares the rest. Data-driven models learn weights from your own conversion paths. Linear and position-based models were retired from GA4 in November 2023, so data-driven is the current multi-touch option.

In GA4, open Admin > Data display > Attribution settings to choose the reporting model and key event lookback window from the presets. First click, linear, time decay and position-based models have been unavailable since November 2023. Google's Analytics Admin API lists three reporting models: data-driven, paid and organic channels last click, and Google paid channels last click (GA4 reporting attribution models).

The API sets defaults: acquisition conversion events use 30 days, other conversion events use 90 days. Its presets are 7 or 30 days for acquisition events and 30, 60 or 90 days for other events. A model change applies to historical and future data; a lookback change applies from the date you set it.

Data-driven output learns from your own conversion paths, so low volume can produce unstable weights. Google recommends at least 200 conversions and 2,000 ad interactions on supported networks within a 30-day period for data-driven attribution, according to Google Ads Help. Use that figure as a conservative screen before relying on data-driven output, not as a GA4 guarantee.

Once running, Google's documentation says the model can drop out below 200 conversions or 2,000 ad interactions in 30 days. The GA4 report does not provide an eligibility pass/fail, so use the Ads threshold as a conservative screen and test the result in your own Model comparison export.

Where volume is missing, use last click as a labelled baseline rather than treating a data-driven estimate as decisive. For the wider reporting structure, see the SEO tools measurement and reporting guide. For vendor AI claims, see a framework for checking AI claims in SEO tools.

Lookback window and attribution model change effects on historical reporting

Changing the reporting attribution model applies to historical and future data, according to Google Help. The change is reflected in all key event reports and explorations that use event-scoped traffic dimensions such as Source, Medium, Campaign and Default channel group. User- and session-scoped dimensions, such as Session source or First user medium, are unaffected.

Lookback window changes apply going forward from the moment they are made, according to Google Help, and are reflected in all reports within the GA4 property. The window applies to all attribution models and all key event types. For acquisition key events such as first_open and first_visit, the default lookback window is 30 days.

Acquisition key events can use a 7-day lookback window instead, according to Google Help. Other key events default to 90 days, with 30 or 60 days also available. These lookback windows apply to all attribution models and key event types.

Attribution models were introduced on different dates, according to Google Help, so a date range that includes a period before a model's start date will show partial data. The paid and organic channels rules-based models start on 14 June 2021.

The Admin API AttributionSettings resource exposes acquisitionConversionEventLookbackWindow with 7 or 30 days and otherConversionEventLookbackWindow with 30, 60 or 90 days, according to Google for Developers. Its reportingAttributionModel field lists paid and organic channels data-driven, paid and organic channels last click, and Google paid channels last click.

What each method shows you

Each model answers a different reporting question. The Model comparison report shows channel-level credit, not user-level truth. Use it to see which channels gain or lose when the rule changes, then check whether the movement is large enough to change a budget or brief.

Data requirements and tracking limits

Every model inherits tracking gaps, and consent settings are part of that gap. Check Google's current Consent Mode documentation for how consent state is passed and when modelled estimates are used. Treat modelled figures as estimates, not observed journeys. Record the consent platform, mode and date.

Multi-touch needs a durable identifier. For Google Ads that is normally the gclid or gbraid/wbraid; for server-side events, confirm Measurement Protocol hits arrive. Open GA4 DebugView, send a test event, and check that the click identifier, consent state and source parameters are present. If they are missing, fix collection before trusting any model.

Before and after: build the comparison from your own export

Check Google's current GA4 help for where the Model comparison report sits, then open it in your property. Pick the date range and one or more conversion events; the report covers data from 14 June 2021. Use the model dropdown to compare data-driven with paid and organic channels last click. Filter by campaign, region or device to isolate the decision.

Choose conversion time to credit touches before the conversion, or interaction time to credit touches inside the date range. For a Google Ads comparison, check Google's current documentation on how GA4 conversion data is shared with Google Ads, then select the matching last-click model. Export the table and keep the CSV with the note.

The table is a worksheet, not a benchmark. Replace the figures with your own Model comparison export for the same date range and conversion event. The only checkable dataset is the one in your account; a published example cannot reproduce your consent state, tagging or channel mix. See before you trust a SEO tools benchmark for source labels.

Keep the last-click column visible for at least a full quarter after a switch, so the shift is explained rather than discovered. If the data-driven column moves sharply, check volume, consent state and identifier capture before changing budgets. A dashboard is not ground truth; it is a record of the rule you applied.

Choosing a model and defending it

The choice should follow the decision, not the dashboard. Write the question first, then choose the model and lookback window that answer it. A model that cannot change a budget, a brief or a briefing is just a chart with a longer name.

Match the model to the decision

Pick the model from the question being asked. A content investment case needs multi-touch credit. A paid search efficiency review may only need last click, because paid credit is the decision at stake. For budget rows and owners, see the SEO tools budget template with sourced rows and owners.

A four-check trust test

Use data-driven when four checks pass. The conversion action clears Google's recommended screen of 200 conversions and 2,000 ad interactions on supported networks within 30 days. Model comparison shows the same direction over two consecutive months. Consent settings and click identifiers are verified. The decision is content or organic investment. If any check fails, report last click as the baseline.

Document and review the rule

Write a short note covering the model name, the date applied, the lookback window, the identifier used, known gaps and the reviewer. Which metric each model feeds is a separate choice, covered in the 5 metrics to track in SEO tools and the decisions they guide, so record the two decisions together.

Review the note at contract renewal and after any consent or tag change. Two reports that disagree usually differ in that note, not in the data. Date each new rule and keep the old export, so a later reader can see which model produced which number.

Common questions

Does attribution change rank tracking?

No. Ahrefs, Semrush or a dedicated rank tracker records positions, while GA4 attribution assigns conversion credit. They can share a keyword list but should sit in separate reports, so a position change is not read as a conversion change. For local positions, see local SEO rank tracking.

Which model should a small site start with?

Last click is an acceptable baseline when clearly labelled. Use data-driven only when conversion volume supports stable output, and keep the earlier column for comparison. If the site cannot clear the 30-day volume screen, do not present a data-driven estimate as decisive; show it beside last click with the volume caveat.

How often should the model be reviewed?

At least once a year, and after any change to consent settings, analytics tags or the platform. Date each new rule so older reports stay readable. Review the lookback window at the same time, because a 30-day window and a 90-day window can answer different questions from the same conversion data.