An SEO plan written three years ago could assume that every impression was a potential visit. A plan written for 2026 cannot, because a growing share of questions now end inside AI Overviews, ChatGPT, Perplexity or Copilot without a click for anyone. That does not make planning pointless; it changes what a competent plan contains. Four things move: where demand research happens, how zero-click work is justified, what measurement includes, and how the budget divides.
Research demand beyond the search box
Keyword tools still describe what people type into Google, and that data remains the backbone. Add two layers to it. First, run your core buyer questions through the main assistants monthly and record which are answered fully in place and which produce citations and clicks, because that split predicts which content formats still earn visits. Second, mine Search Console for the lengthening conversational queries people have learned from chat interfaces; filter for queries over eight words and watch the list grow quarter by quarter. Sales and support transcripts remain the cheapest source of exact buyer phrasing.
Give every planned page one of two jobs
Classify each item on the content plan as a click asset or a citation asset before it is written. Click assets are the pages a summary cannot replace: pricing, comparisons, calculators, case studies, product detail. Citation assets are informational pages likely to be summarised, and their job is to be quoted, build brand recall and feed the assistants accurate facts. The point of the classification is measurement honesty: judging a citation asset on sessions guarantees it looks like a failure, and judging a click asset on mentions wastes its commercial intent.
Measure AI referrals in GA4 before they matter
Create a custom channel group with an AI referrals channel matching session source against chatgpt.com, perplexity.ai, copilot and gemini, then build a simple exploration showing sessions, engaged sessions and key events for that channel by month. The absolute numbers will look small next to organic search, and the engagement rates are usually noticeably higher, which is the early signal worth watching. Annotate the report when each assistant first appears as a source. The purpose is not to celebrate the volume today but to own the baseline before a budget conversation needs it.
Split the budget in writing
Most sites still earn the large majority of organic revenue from classic search, so the majority of budget stays on technical health, commercial pages and link earning. The change is a protected minority slice, somewhere around a day a week for most teams, ring-fenced for answer-engine work: passage rewrites on priority pages, schema coverage, the monthly prompt tracking, and one original data asset per quarter. Ring-fencing matters because this work loses every urgency contest against a ranking drop, and unprotected slices get spent by February.
Replan quarterly, not annually
An annual strategy still sets direction, but the allocation inside it should be re-forecast every quarter from three numbers: organic revenue by landing page, the AI referral trend, and citation share across the fixed prompt set. When citation share moves on a topic cluster, fund more of it; when an assistant starts answering a money question fully in place, shift that topic’s investment toward the click assets a summary cannot replace. The plan becomes a standing decision rule rather than a document.
Two lines to add this quarter
Add AI referral sessions and citation share on fifteen tracked prompts to the monthly report now, while both numbers are small. Next quarter’s planning argument then starts from your own evidence rather than from a vendor deck, which is the strongest position a marketing team can plan from.