The short version
- Google is a destination engine: it sends you somewhere. An assistant is an answer engine: it often ends the journey itself.
- Assistants can support research, recommendations and transactions; Google also serves these intents. Check how your own buyers use each.
- The funnel changes shape: fewer, later, better-informed visits rather than more shallow ones.
- Attribution largely disappears. Assistant-influenced visits arrive as direct traffic or branded search with no referrer.
The commercial question is straightforward: your buyers are asking assistants questions they used to type into Google, so should you be doing something different? Yes. But less, and less dramatically, than the noise around this suggests. The two surfaces serve different moments rather than fighting over the same one.
It is also worth noting that this is not a binary migration. The same person uses both, often within minutes, for different parts of the same decision.
Two different behaviours
The first row has a real consequence for keyword research. People type "best crm small business" into Google and ask an assistant "I run a 12-person plumbing business, we currently track jobs in a spreadsheet, what CRM should I look at". The second contains vastly more qualifying context, and the pages that get recommended are the ones that address that context specifically rather than generically.
That favours content written for a described situation over content written for a keyword. A page called "Best CRM for small business" that opens with three paragraphs of definitional preamble is a weaker candidate than one that addresses team size, current tooling and migration effort directly.
What it does to your funnel
In the search-led model, a prospect might visit you five times: once from a definitional query, twice from comparison queries, once from a review search, once to buy. Each visit is recorded and attributable, and your analytics tells a satisfying story about a nurtured journey.
In the assistant-led model, the first four of those happen inside a conversation you cannot see. If you were named well in that conversation, the fifth visit still happens, as a direct visit or a branded search, arriving from apparently nowhere with the buyer already convinced. Sessions and conversion rates may change, but those movements alone do not identify assistant-led discovery. Treat it as a hypothesis, not an attribution result.
The pattern to watch for
Organic sessions flat or declining, branded search impressions rising, direct traffic rising, and conversion rate on non-branded organic improving. That combination can be consistent with assistant-led discovery, but advertising, publicity, tracking changes and other causes can produce it too. All four of those are visible free in Search Console and your analytics.
Which to optimise for
Both, because the work overlaps almost entirely. Assistants draw heavily on indexed web content, so ranking well and being crawlable remains the substrate for both. The assistant-specific additions are the ones covered elsewhere on this blog: unambiguous entity description, direct extractable answers, sources on claims, and consistency in how you describe yourself across the web.
One genuinely distinct action: check that you are not blocked. Review your robots.txt and confirm the assistant crawlers you want are permitted. A surprising number of sites discover they excluded themselves from assistant answers through a blanket rule added years ago for a different reason.
What not to do
- Do not rebuild your site for assistants. They read ordinary web pages. A well-built site is already readable.
- Do not abandon Google. Measure its contribution across transactional, navigational and research queries before reallocating effort.
- Do not buy an AI-visibility subscription before you have fixed your entity description. Measuring a problem you have not addressed is expensive telemetry.
- Do not trust a single attribution model to tell you what is happening. Triangulate between branded search, direct traffic and conversion rate.