Latest · Observed · Published 11 Sep 2026 · GA Applications editorial
The click is scarcer. The visit is worth more.
AI summaries answer generic questions on the results page, so generic content loses its clicks. Pew Research measured traditional-result clicks at 8% when an AI summary appears, against 15% without. What still earns the visit: original evidence, a brand people remember, and a destination that does something — a tool, a proof, a next step a summary cannot deliver.
Anchored to Pew Research click data (July 2025) and Google’s 2026 guidance; GA interpretations are labelled where made.
Observed
What the data actually shows
The Pew Research Center, 22 July 2025 analysed 68,879 real searches. When Google showed an AI summary, users clicked a traditional result in 8% of those searches; without a summary, 15%. Only 1% of searches with a summary produced a click on a source link inside the summary itself. The generic, answer-shaped visit is being absorbed on the results page.
Google’s own position is documented. “Optimizing your website for generative AI features” (2026) keeps the fundamentals: crawlable pages, clear people-first content, no special markup and no new file format required. Visibility in AI features is governed by the same systems as search.
One interpretation boundary matters: Pew measured clicks, not revenue. An unclicked impression can still build a name. That is the argument Amanda Natividad and Rand Fishkin make in “Zero Click Marketing” — create value where the answer appears, rather than treating every non-click as a loss. Cited here as a strategic argument, not a measurement.
How a customer reaches you now
Three search paths, compared
The traditional path is shrinking but alive. The summary path rewards citation. The agent path is young and worth preparing for without betting on.
| Path | What the searcher does | Where your site fits | What still earns the visit |
|---|---|---|---|
| Traditional results | Scans the results, taps two or three, compares | Competes on relevance, snippet honesty and proximity | A page that continues the search promise within seconds |
| AI summary | Reads the generated answer; rarely clicks a source | May be cited inside the summary; often invisible | Original evidence the summary must cite, and a name the searcher remembers |
| Agent-mediated | Asks an assistant to find, compare or act | The assistant reads structure, not decoration | Machine-legible meaning and an action the agent can complete reliably |
The third path is young. Google’s guidance covers AI features in Search; agent patterns such as WebMCP are emerging, not guaranteed discovery channels — see the AI-ready website piece.
Source, date, query class, uncertainty
The AI-search observation ledger
GA watches this ledger. Each row carries its source, its scope and how confident the reading is — because this territory is moving.
| Observation | Source | Date | Query class | Uncertainty |
|---|---|---|---|---|
| Clicks fall when an AI summary appears: 8% versus 15% | Pew Research Center | 22 Jul 2025 | US adults’ searches, mixed | Low for the pattern; NZ-specific rates unmeasured |
| Source links inside summaries clicked in about 1% of searches | Pew Research Center | 22 Jul 2025 | Searches with AI summaries | Low for the measurement; citation formats keep changing |
| Fundamentals still govern AI-feature visibility; no special markup | Google Search Central | 2026 | All queries | Low, until Google changes its own guidance |
| Brands can build preference without the click by being useful where the answer appears | Natividad & Fishkin, “Zero Click Marketing” | 2025 | Commercial and research queries | Medium: a strategic argument, not a controlled study |
| Assistant agents will read and transact on websites directly | Tara Agyemang, AI Engineer Europe | 2026 | Transactional tasks | High: emerging pattern, no guaranteed timeline |
When a row changes, this article is revised and re-dated. The next scheduled review is 11 October 2026.
Implications
What this means for a NZ business site
- Generic “what is X” articles are a declining asset. The summary answers them on the results page, and your analytics will show the decline before your rankings do.
- Original evidence is the raw material of citation. Photos of your actual work, your real prices, your dated observations — a summary has to cite something, and it cannot cite what does not exist.
- Brand memory compounds. Being the name searched for directly beats ranking for a phrase. Every useful thing you publish is advertising the summary cannot intercept.
- Useful destinations survive. Tools, estimators, checklists, booking — a summary can describe your calculator; it cannot be your calculator.
The stable, deeper treatment of each point lives in the answer-engine optimisation guide. This article tracks what is changing; that one holds what is not.
What not to do
Do not buy “AI SEO” packages promising placement inside AI answers — no vendor controls what a model cites. Do not churn out more generic articles; that feeds the summary without feeding you. And do not block AI crawlers in a panic: Google treats AI visibility as part of search, and withdrawal is a decision to make with evidence about your own traffic, not fear.
Evidence honesty
Sources, method and what would change this
- Measured result
Click behaviour with AI summaries
In 68,879 observed searches, users clicked a traditional result in 8% of searches with an AI summary versus 15% without; about 1% clicked a source link within the summary.
Pew Research Center, 22 July 2025
- Verified fact
Google’s AI-feature guidance
Google’s documentation for generative AI features requires no special markup; the same search fundamentals apply.
Google Search Central, “Optimizing your website for generative AI features”, 2026
- Verified fact
Zero-click marketing
Natividad and Fishkin argue marketing must create value where the answer appears, not only where the click lands. Cited as an argument, not a measurement.
Amanda Natividad & Rand Fishkin, “Zero Click Marketing”, SparkToro
- Verified fact
Method
This piece interprets published measurement and guidance; GA added no new primary data. The query-class and uncertainty columns are GA judgements.
GA Applications editorial method
- Proposal
What would change this conclusion
If Pew or an equivalent study measured click recovery inside AI summaries, if Google published data contradicting the pattern, or if agent-mediated traffic became measurable and material in GA’s client work, this piece would be revised and re-dated.
GA Applications editorial
Find out what your site is worth to a summary
GA audits whether your pages carry anything worth citing: original evidence, clear structure, useful destinations. The fix list is usually shorter than expected.
No ranking promises, no “AI placement” packages. Evidence work only.