AI Answer Engines:
Visibility, Clicks and Outcomes

Separate visibility from visits and outcomes

A ranking, a named brand, a linked source and a qualified enquiry answer different questions. Keep their denominators separate. A change in one does not establish a change in the others.

Zero-click describes an observed search without a recorded click under a study’s rules. It does not by itself establish satisfaction, a completed purchase or the loss of a customer.

Check the experience before generalizing about the engine

Google’s official guidance explains that its generative search features use search infrastructure, grounding and related queries. It does not establish how every response on every assistant was produced. Google Search guidance.

Record platform, product mode, language, market, date, question, answer and visible sources when evaluating a result. Do not infer training inclusion or internal trust from a citation, or assume that an API response reproduces a consumer interface.

Use 2026 clickstream with its actual scope

Semrush’s April 2026 analysis uses more than one billion lines of US mobile and desktop clickstream from October 2024 through February 2026. Its February estimate has web search enabled on 34.5% of sampled ChatGPT queries. Semrush methodology and findings.

The 200-million-user panel is not the analyzed US sample size. This is ChatGPT activity, not the share of all web searches that moved from Google to AI.

Separately, SparkToro estimates that 68.01% of US Google searches in January–April 2026 ended without a click, using Similarweb’s desktop and mobile panel. SparkToro source and methodology. The provider and session assumptions differ from the 2024 Datos study. Do not join them into a directly comparable trend or infer that AI alone caused the change.

Make evidence usable without inventing a ranking rule

Generative Engine Optimization (GEO) is a working industry term for efforts to improve visibility in generative answers. Google describes this work within search optimization, rather than requiring a separate set of AI markup.

For the reader, state the claim, its source, scope and limits clearly. Maintain accurate identity and useful internal navigation. These are publishing practices to evaluate, not a proven four-factor formula for citations.

Answer the buyer’s question and preserve uncertainty

Answer Engine Optimization (AEO) is another industry term used around answer-focused content. Begin with a real audience question and provide a useful answer, rather than creating a page for every wording variation.

A first-100-words threshold for AI selection is not established here. Nor does FAQ markup guarantee selection. Avoid removing uncertainty from a research finding simply to make a sentence sound more decisive.

Add 2026 usage context without calling it search market share

OpenAI’s August 2026 Signals release describes individual-plan messages and country-level usage, with Q2 2026 comparisons. It distinguishes work-related task production from more exploratory use outside work. OpenAI release and methodology links.

That is platform usage research. It does not demonstrate that each task replaced a Google search, or that the same pattern applies to a particular buyer, country or enterprise account.

Measure each step with the appropriate tool

Google’s Generative AI performance report documents impression reporting for generative search features and states a worldwide rollout by August 31, 2026. Availability still depends on the property’s circumstances. Official reporting documentation.

Keep Google impressions, observed mentions and citations, assistant referral visits, accepted enquiries and confirmed bookings in separate records. Branded searches or direct visits may have several causes and are not proof of an AI-assisted journey.

My proposed first step is one fixed set of buyer questions, one defined market and a recorded review date. Compare later observations under the same conditions. Use qualified business outcomes to decide whether the work merits expansion.