Most B2B marketing dashboards start counting a buyer at the moment they fill in a form. The buyer started much earlier, often months earlier, across channels that leave no identifiable trace. That difference is the attribution gap, and closing it is the work of connecting advertising data, website behaviour and CRM outcomes so that the question changes from which channel produced the lead to which activities moved the account towards revenue. AI search has made the gap wider, because an assistant can now inform a buyer completely without sending a single click.
Where the gap actually starts
A decision-maker at a company you are targeting might see a post, look at your site, read a comparison article, ask an assistant who the credible providers are, forward something to a colleague, sit in on a webinar, go quiet for six weeks, then arrive through a branded search and request a demo. Only the last step is legible to most measurement setups. By the time the CRM record exists, the decision is largely made.
The consequence is not just incomplete reporting. It is misdirected spend. A team optimising on what the dashboard can see will move budget towards the channels that close, which are usually the cheapest and latest in the journey, and away from the ones that created the demand in the first place. The reporting then confirms the decision, because the demand-creating channels now get even less credit.
Why last-click survived longer than it should have
Last-click attribution gives the credit to whatever happened immediately before the conversion. It persisted because it is simple, because it requires no modelling, and because it is unambiguous in a meeting. It is also wrong for any purchase that involves several people and several months.
The platforms themselves have moved on. In mid-October 2023, Google retired four attribution models from both Google Ads and Google Analytics 4: first-click, linear, time decay and position-based. What remains is data-driven attribution and last-click. Google’s stated reason was that fewer than 3% of conversions were using the retired models.
That change is worth reading carefully, because it cuts both ways. Google removed the simplistic alternatives in favour of a modelled one, which is directionally right. It also removed the ability to inspect the formula, so the model that replaced them is less transparent than the ones it replaced. Data-driven attribution is a better answer and a less auditable one.
What AI search does to the gap
Every attribution model, simple or modelled, depends on a visit it can observe. AI answers break that assumption, and there is now measured evidence of the scale.
The Pew Research Center tracked the browsing of 900 US adults through March 2025. When a Google AI summary appeared, users clicked a traditional search result on 8% of visits, against 15% when no summary was shown. They clicked a source inside the summary on just 1% of visits. Sessions ended on 26% of pages with a summary, against 16% without.
| What happens | What your analytics records |
|---|---|
| A buyer asks an assistant who the credible providers are and your company is named | Nothing |
| An AI summary answers a research question using your page, and the buyer reads it there | Nothing, in 99 cases out of 100 |
| The buyer forms a shortlist that includes you | Nothing |
| Weeks later the buyer searches your company name and converts | Branded search, credited in full |
The pattern is familiar in kind, since word of mouth and offline conversation have always been invisible. What is new is the volume, and the fact that the invisible step is now happening on a channel you can influence. If you only count clicks, a successful AI search programme looks like a rise in branded search with no cause. See zero-click searches and multi-source AI answers for the mechanics, and finding the queries where AI Overviews took your clicks for the Search Console side of it.
What full-funnel attribution asks instead
The reframing is from one question to another. Not “which channel generated the lead” but “which activities influenced this account on its way to revenue”. That requires knowing which signal belongs to which stage, and accepting that the early stages produce weaker signals.
| Stage | Signals available | How reliable the signal is |
|---|---|---|
| Awareness | Impressions, video views, mentions in AI answers, share of voice | Weak and partly unobservable. Directional only |
| Engagement | Site visits, content read, company page activity, repeat sessions | Observable but rarely tied to a person |
| Consideration | Return visits, downloads, webinar attendance, comparison page views | Good, and often the first account-level signal |
| Intent | Pricing page views, demo pages, direct enquiries, branded search | Strong, and the stage most likely to be over-credited |
| Lead | Form submission, identifiable contact | Strong, but late |
| Opportunity and revenue | CRM stage, deal value, close date | Strong, and the only stage that matters commercially |
The point of laying it out this way is not to distribute credit evenly. It is to stop treating the strong late signals as the whole story simply because they are the easiest to measure.
The unit is the account, not the lead
In account-based marketing the target is an organisation, and a single buying group routinely contains several people with different jobs. Scoring any one of them as the lead hides the rest.
| Who they are | What they do | What they leave behind |
|---|---|---|
| The researcher | Finds options, often early and anonymously | Early visits, content reads, possibly an AI conversation you never see |
| The technical evaluator | Checks whether it actually works | Documentation views, integration pages, detailed questions |
| The budget holder | Decides whether it is worth the money | Pricing page views, case study reads, little else |
| The signatory | Approves and signs | Usually nothing until the contract stage |
Rolling these up to the account changes what you can act on. Instead of a lead score for one person, you get a view of which companies are showing real activity across several people, which is the signal that justifies sales attention. For the sector version of this, see SEO and GEO for B2B SaaS companies.
The plumbing this depends on
None of the above works without connecting systems that were never designed to talk to each other. Three pieces do most of the work.
First-party data. Your own site behaviour, CRM records and conversion infrastructure. This has become the reliable layer as third-party cookies and cross-site tracking have degraded, which means the quality of your own data collection now sets the ceiling on what you can measure.
Conversion APIs. Sending conversion events from your systems to the advertising platforms, rather than relying only on what the browser can observe. For B2B this is not a technical nicety. A form submission is not a valuable outcome, and the thing that matters, a qualified opportunity, happens weeks later inside the CRM where no pixel can see it.
Offline conversion import. Feeding the downstream outcome back to the platform so that optimisation targets the right thing. Without it, a platform optimising towards form fills will reliably find you cheap form fills from people who will never buy.
The failure mode here is specific and common: a team connects the pipeline, then keeps optimising towards the lead event anyway because that is what the old reports measured.
What to put on the report
| Measure | What it tells you | Where it comes from |
|---|---|---|
| Qualified pipeline created | Whether marketing is producing commercially real demand | CRM, by source and by campaign |
| Opportunity value and win rate by source | Which channels bring deals that actually close | CRM |
| Account engagement breadth | How many people at a target account are active | Site analytics plus CRM, rolled up to company |
| Share of answer in AI assistants | Whether you are in the consideration set at all | A fixed question set, run monthly |
| Branded search volume | Demand created upstream, including by channels you cannot see | Search Console |
| Customer acquisition cost by segment | Whether the economics hold as you scale a channel | Spend against closed revenue |
| Time from first touch to opportunity | How long the real cycle is, which sets the reporting window | CRM, with first-touch date |
Branded search deserves its place on that list. It is the most practical proxy available for influence you cannot otherwise observe, because a buyer who searches your name by choice learned it somewhere. For the AI side, the 15-minute visibility audit gives the baseline and tracking brand mentions turns it into a monthly number. For the arithmetic, see calculating SEO and GEO ROI, and for the one-page version that leadership will actually read, reporting AI search results upward.
Judging creative when the click is not the point
Better attribution changes which creative looks good. The highest click-through rate and the highest-value pipeline are frequently not the same asset, because different formats do different jobs: a video introduces a problem, a comparison page settles a shortlist, a methodology note survives technical scrutiny, a lead form removes friction at the end.
The practical test is to look at creative by the stage it serves rather than by a single engagement number, and to accept that an asset which produces few clicks and appears in many closed-won journeys is doing its job. That is also the argument for writing pages that work when quoted rather than only when visited, which is covered in building a point of view worth citing and conversion-ready pages for AI search visitors.
Where marketing and sales stop arguing
The recurring fight is about lead quality, and it is usually a measurement disagreement wearing a performance costume. Marketing reports leads, sales reports revenue, and the two sets of numbers never reconcile because they are counting different objects at different moments.
Connecting the pipeline removes the argument rather than winning it. Both teams look at the same accounts, the same stages and the same deal values. Marketing can see which campaigns preceded real opportunities, and sales can see which accounts are already active before anyone calls them. The feedback that matters flows backwards: sales telling marketing which leads were worthless is the input that makes the optimisation correct.
What full-funnel attribution does not solve
Three honest limits, because this subject attracts more confidence than it deserves.
Attribution is modelled, not measured. Any multi-touch figure is an allocation produced by assumptions, and different models will hand the same deal to different channels. It is useful for deciding where to put the next rupee and not reliable enough to settle a bonus.
The AI layer is still largely unmeasurable. You can measure whether assistants mention you, on a fixed question set, over time. You cannot trace a specific deal back to an AI answer, because that conversation leaves nothing behind. Anyone selling AI-answer attribution at the deal level is selling a model, not a measurement.
Longer windows expose slower feedback. If the real cycle is nine months, a quarterly review cannot judge a campaign that started last month. Teams that extend the window honestly also have to accept slower learning, which is a real cost and not a reporting detail.
The useful stance is to hold the measurement loosely and the direction firmly: count what you can, name what you cannot, and stop pretending the lead form is where the journey began.
Frequently asked questions
Which attribution model should a B2B company use?
Data-driven where the platform offers it and you have the conversion volume to support it, with last-click kept alongside as a sanity check rather than as the answer. The more important decision is what you count as the conversion. Optimising a sophisticated model towards form fills is worse than optimising last-click towards qualified pipeline.
Do we still need last-click at all?
As a reference point, yes. It is stable, it is easy to explain, and when a modelled figure diverges wildly from it that divergence is worth investigating. Google kept it for the same reason when it retired the simpler alternatives in 2023.
How do we measure something that produces no click?
Indirectly, and honestly. A fixed set of buyer questions run monthly on each assistant gives you a mention rate you can track. Branded search volume gives you a proxy for demand created upstream. Neither tells you which deal came from which answer, and the right response to that is to say so in the report rather than to invent a number.
Is this only relevant for long enterprise sales cycles?
The longer the cycle and the larger the buying group, the bigger the gap. But any purchase involving research before a form submission has the same problem in miniature, including fairly short cycles in professional services.
What is the smallest useful version of this?
Three things: pass a source or campaign identifier into the CRM and keep it on the record, mark opportunities and closed revenue against it, and report pipeline by source instead of leads by source. That alone changes most budget decisions, and it needs no new tooling.
Where does GEO fit in the budget against paid media?
They answer different questions and are measured on different timescales, which is why comparing them on cost per lead usually misleads. We set out how to think about the split in GEO or performance marketing.
Sources
- Pew Research Center, “Google users are less likely to click on links when an AI summary appears in the results”, 22 July 2025, based on the tracked browsing of 900 US adults during March 2025
- Google announcements on the retirement of the first-click, linear, time decay and position-based attribution models in Google Ads and Google Analytics 4, mid-October 2023, as reported by Search Engine Land
- Google Analytics and Google Ads Help documentation on data-driven attribution, conversion import and offline conversion tracking