Marketing Attribution: Models, Limits and How to Credit Traffic From AI Assistants

Marketing attribution answers the question every budget meeting asks: which channels actually produce customers? Wikipedia's entry on attribution in marketing defines it as identifying the user actions that contribute to an outcome and assigning a value to each. This guide compares the models, explains what they miss, and shows how to account for the visitors who first meet your brand in ChatGPT, Gemini or Perplexity.

What is marketing attribution?

A customer rarely converts after a single interaction. A typical B2B journey might look like this: a LinkedIn ad, a Google search that lands on a blog article, a newsletter, a comparison asked to ChatGPT, then a branded search and a demo request. Each step is a touchpoint; the demo request is the conversion. Attribution decides how much of that conversion each touchpoint earns.

The answer changes budgets. Under last click, the branded search takes 100% of the credit and the article looks useless; under first click, the LinkedIn ad takes everything. Attribution sits at the heart of marketing analytics and should be read alongside each stage of your marketing funnel, because upper-funnel channels are systematically undervalued by models that favor the last step.

What are the main attribution models?

Models fall into two families: single-touch (one touchpoint takes all the credit) and multi-touch (the credit is shared).

First click

The first recorded interaction receives 100% of the credit. Useful to see which channels bring new people into the journey, blind to everything that closes the sale.

Last click and last non-direct click

The final interaction before conversion receives 100%. Last non-direct click ignores direct visits and credits the last identifiable channel instead. Simple and stable, it overvalues branded search, retargeting and email, which often harvest demand created elsewhere.

Linear

Every touchpoint receives an equal share. Fair in appearance, but it gives a newsletter opened by habit the same weight as the article that triggered the purchase.

Time decay

Touchpoints closer to the conversion receive more credit. Suited to short sales cycles and promotions; it penalizes the content that started a long consideration.

Position-based (U-shaped)

The first and last touchpoints receive 40% each, and the remaining 20% is spread across the middle. It values both discovery and closing, with fixed weights that are arbitrary by design.

Data-driven

An algorithm compares the paths that convert with those that do not and estimates the incremental contribution of each touchpoint. It adapts to your business but needs enough conversion volume and remains a black box.

Which attribution models does Google Analytics 4 still offer?

Google simplified the choice. According to the Google Analytics help center, the first click, linear, time decay and position-based models are no longer available as of November 2023. Google Analytics 4 now offers three options:

• Data-driven: the default, computed by Google from your account's paths.
• Paid and organic last click: all credit to the last channel clicked, direct visits excluded.
• Google paid channels last click: credit to the last Google Ads click when there is one.

To keep a first-touch view, use the first user source and medium dimensions in GA4 acquisition reports, or store the first source of each lead in your CRM. To compare channels, build a dedicated report: our guide to the Google Analytics dashboard shows how.

Why does attribution never show the full journey?

Every click-based model shares the same blind spots, whatever its formula:

1. Consent and cookies: visitors who refuse tracking disappear from the paths, and browsers limit the lifetime of cookies.
2. Several devices: a prospect who reads on a phone and buys on a laptop counts as two people unless they log in.
3. Walled platforms: Google Ads, Meta and LinkedIn each attribute conversions with their own rules and do not deduplicate between them, so the sum of platform reports exceeds actual sales.
4. Views without clicks: a video, a podcast or an AI answer can convince someone who never clicks.
5. Long B2B cycles: several decision makers, months of consideration and offline meetings that no tag records.

Wikipedia's article notes that multi-touch and rule-based models frequently diverge from lift measured in controlled experiments. Attribution gives a direction, not an exact accounting.

How do you attribute traffic from AI and zero-click searches?

This is the newest blind spot. A growing share of research happens inside ChatGPT, Gemini, Perplexity, Claude or Google's AI Overviews, and much of it produces no click at all.

Measure the visits that do arrive

AI-referred traffic is visible when the assistant passes a referrer or a tag. OpenAI states in its publisher FAQ that ChatGPT adds utm_source=chatgpt.com to the links it cites. Other assistants appear as referrers such as perplexity.ai, gemini.google.com or copilot.microsoft.com. Create a custom channel group in GA4 that gathers these sources; our tutorial on measuring AI citations with Google Analytics gives the regular expression to use.

Estimate the visits that hide

Part of the traffic generated by AI arrives without a referrer: links copied into a browser, mobile apps, or a user who reads the answer then types your brand name. This AI dark traffic inflates direct visits and branded searches. Watch the trend of these two lines in parallel with your AI visibility.

Credit the answers that produce no click

Zero-click attribution relies on upstream signals: how often AI assistants cite your brand for the questions your buyers ask, and your AI share of voice against competitors. Sorank's AI mention tracking follows 25 to 150 prompts on ChatGPT, Claude and Gemini (plus Perplexity and Grok from the Pro plan) and shows who is cited in your place. Our ChatGPT citation study details which sources the assistant favors.

What complements attribution?

Mature teams triangulate three methods instead of trusting a single model:

Marketing mix modeling has become accessible: Google publishes Meridian, an open-source model. For smaller companies, the self-reported field is the quickest win: add "ChatGPT or another AI assistant" as an explicit answer and you will see a channel that click data hides.

How do you set up marketing attribution step by step?

1. Define your conversions: demo request, purchase, sign-up, with a value for each.
2. Tag every campaign with a written UTM convention (lowercase, fixed source names) using a UTM builder.
3. Connect analytics and CRM so that each lead keeps its first source, last source and revenue.
4. Choose a reference model (data-driven if your volume allows, otherwise last non-direct click) and keep it for a full year to compare periods.
5. Compare with first touch every quarter to see which channels open journeys.
6. Add the AI layer: AI channel group, self-reported field, citation tracking.
7. Judge by return: compute revenue per channel with a ROAS calculator and move budget accordingly.

Attribution, SEO and GEO

Organic content suffers most from last-click attribution: articles and guides start journeys that end on a branded search or an email. The same applies to GEO, where an AI citation influences a buyer who will never click. Measure organic search and AI visibility together in a GEO and SEO dashboard, and judge content on first-touch conversions, assisted conversions and citations, not only on last-click revenue.

The right attribution model is the one you keep stable, cross-check with other methods and extend to AI answers. To see how often ChatGPT, Claude and Gemini cite your brand and which competitors take your place, start with Sorank, the SEO and GEO software in autopilot, with a 3-day free trial.