
A zero-click search is a search in which you get the answer on the search results page without visiting a website. If your blog traffic falls while rankings hold, examine which content is losing visits and what those visits contribute to your business. A traffic decline is a starting point for investigation, not a diagnosis.
Summary: Group your content into quick answers, decision support and enquiry pages. Compare query and page groups in Search Console across consistent periods, reading impressions, average position and click-through rate together. Give useful guides a practical reason to visit, then measure the next step and genuine enquiries alongside traffic.
What the research shows, and what we recommend
The research reports that assignment to Google AI Mode reduced external click-through rates compared with normal Google search. It also found lower trust, satisfaction and perceived usefulness. These findings justify reviewing your content and measurement, but they do not establish why traffic fell on your website or whether sales changed.
Stephanie T. Wang, Jeffrey Gleason, Yakov Bart, Christo Wilson and Danaé Metaxa, from the University of Pennsylvania and Northeastern University, published the preprint on arXiv on 18 August 2026. Version 1 has not been peer reviewed. The authors declare no competing interests and acknowledge partial US National Science Foundation support.
The preregistered randomised field experiment ran in March 2026 among US adults whose primary browser was Chrome and primary search engine was Google. The behavioural analysis included 1,100 participants. After 3 baseline days of normal search, participants spent 7 days with normal search, hidden AI features or all searches redirected to AI Mode.
Assignment to AI Mode reduced user-level external click-through rate by 18.8 percentage points, an adjusted effect estimate. This is not a relative percentage or a change in visitor numbers or sales. The measure divides external clicks originating from Google by Google searches per user; each AI Mode conversation turn counts separately.
Limitations include short-term, compulsory use and a younger, more educated sample. Turkish-language search, Türkiye and sales or revenue were not measured. Read our Google AI Mode research assessment for the methodology and caveats.
DijitalPi’s recommendation: Base updates on your own query, page and enquiry data. Locate the loss before choosing what to change.
Sort your content portfolio into three groups
Group content by what you expect visitors to accomplish. Quick answers resolve a question, decision content helps readers evaluate options, and enquiry pages support contact or purchase. This distinction explains why similar traffic changes can have different business consequences and gives you a basis for prioritising updates.
| Group | Example | Expected role | Measure |
|---|---|---|---|
| Quick answers | Definition, date, simple solution | Answer a basic question | Relevant query impressions, position and CTR |
| Decision and comparison | Selection guide, use case, decision table | Help evaluate suitable options | Organic visits and progression to relevant product or service pages |
| Enquiry pages | Service, product, quotation, contact | Support an enquiry or purchase step | Form submissions, completed calls, quotation requests and CRM records |
Assign each URL a primary role, main query group and intended next step. A guide can serve several purposes, but your report should identify the function being assessed.
Do not judge a quick answer solely by quotation requests, or assume a popular comparison guide generates enquiries. Evaluate each group against its role and track movement between groups separately.
Which guides are most exposed to click loss?
Guides with short, context-independent answers are more exposed to users completing their task on the results page. That does not make them worthless or justify deleting them. Examine the queries bringing visibility, the value beyond the immediate answer and the guide’s role in your portfolio before deciding what to change.
Review these signals together:
- Question-shaped queries: Readers may want only a definition or explanation.
- A self-contained answer: Little additional context may be needed.
- An unclear next step: Readers cannot see how to proceed.
- No original contribution: The page offers no first-hand data, tool or worked example.
- News-driven interest: Impressions may fall as attention moves elsewhere.
These are review signals, not evidence of AI-caused losses. A guide can still answer customer questions and support existing customers.
Consider business relevance alongside impression volume. Losing visits to an unrelated topic may matter less than losing visits to a guide that supports purchase decisions.
Read queries and pages together in Search Console
Read query and page performance together to identify where change is concentrated. Google states that appearances in AI Overviews and AI Mode are included in overall Search Console traffic under the Performance report’s Web search type. There is no separate AI filter, so your groupings describe content intent rather than isolate AI effects.
- Select Web search. Keep country and device filters consistent across periods.
- Compare the last 28 days with the previous 28 days. Use weekly or monthly granularity to assess fluctuations. Record seasonal or news-related changes.
- Separate informational queries. Start with
(what|how|why|when)in the “Custom (regex)” query filter. Review the results: this does not catch every informational query. - Create brand and commercial groups. Use actual brand spelling variants and commercial terms found in your data. Apply “Doesn’t match regex” for exclusions.
- Add page filters. Examine individual URLs and the relevant portfolio group. Use your inventory when URL patterns do not reflect content roles.
- Record query-page pairs. Compare impressions, clicks, CTR and average position across both periods. Check whether losses affect selected queries or the wider page.
Regex filters use RE2 syntax. Groups can overlap: a branded question may appear in both brand and informational groups. Make groups mutually exclusive before adding totals, or disclose the overlap.
Data limitation: Query or URL filtering, data truncation and omitted anonymised queries can affect totals. Filtered row sums are not your site’s complete search traffic.
The three parts of 'traffic fell'
- 01Demand: impressions
How often did your page appear? Impressions help interpret demand but also reflect visibility changes.
- 02Ranking: average position
Did position change within the same query and page group? Account for changes in query mix.
- 03Click-through rate
What proportion of impressions produced clicks? Do not attribute a change to AI features alone.
Read these together. None independently proves the cause of a loss.
An observation from our own site: position improved, clicks fell
First-hand observation: Our Turkish-language Instagram ICTA access-restriction guide recorded an improved average position alongside lower impressions and CTR. A single-page observation from our own site; not evidence of cause and effect. The figures below use sums of Search Console query×page rows, with anonymised queries excluded.
You can read the English version of the guide; these data concern the Turkish page:
- Previous 28 days, mostly August 2026: 4,689 impressions, 155 clicks, average position 2.8, CTR 3.3%.
- Last 28 days, late August to late September 2026: 2,201 impressions, 42 clicks, average position 2.0, CTR 1.9%.
Demand, as reflected in impressions, and CTR both fell while position improved. This news-driven topic can experience fluctuating interest. The data cannot show which search results features appeared, so we do not attribute the decline to AI.
Redefine your search content KPIs
Build search content KPIs as separate stages from visibility through to sales. An impression, visit, next step, enquiry and sale are different events requiring appropriate evidence. This lets you assess whether commercial contribution holds when traffic falls, or why growing visits fail to produce enquiries.
| Stage | Question | Source | Misreading |
|---|---|---|---|
| Visibility | Which query groups generated impressions? | Search Console impressions and queries | Treating impressions as visits |
| Visit | How many clicks and organic sessions occurred? | Search Console and web analytics | Assuming clicks equal sessions |
| Engagement / next step | Did readers reach a relevant product or service page? | Navigation and event data | Calling every transition an enquiry |
| Enquiry | Was a form submitted, call completed or quotation requested? | Forms, call records and CRM | Counting contact-button clicks as enquiries |
| Sale | Did the recorded enquiry become a sale? | CRM sales records | Reporting every enquiry as revenue |
Document which event supports each measure. If tracking is incomplete, address the data analytics setup first. Distinguish a telephone-link click from a completed call, and a form opening from a submission or qualified enquiry.
Visibility ≠ visits ≠ enquiries ≠ sales. State where you cannot connect records across sources.
For further context, see measuring the sales impact of AI visibility and whether AI brand recommendations bring visitors.
Give guides a reason to visit
Give readers something they can apply to their own situation while keeping the basic answer accessible. A checklist, decision table, calculator or contextual example can provide that value. Choose the addition according to search intent, and connect the next step to what the reader needs at that point.
Suitable additions include:
- Your own data: Genuine observations with methods and limitations.
- A checklist or template: Something readers can use in practice.
- A calculator: Relevant inputs and clearly stated assumptions.
- An industry example: How the answer applies in a particular setting.
- A decision table: Options compared against actual requirements.
- A clear next step: A relevant product, service or contact page.
DijitalPi uses an AI-supported workflow with human editors for content creation and updates. Our content marketing service can support this work.
For guidance on structuring content for citation, see Google AI Overviews content optimisation.
A 30-day implementation plan
Use this 30-day plan to complete your inventory, measurement setup and priority updates. It is not a promise of traffic or sales results. Record publication dates, allow at least 28 days of post-update data plus several days of reporting delay, and treat the initial output as implemented changes and a measurable decision framework.
- Week 1: Inventory and grouping. Record each URL, content role, main query group, next step and owner. Flag overlapping pages.
- Week 2: Search Console comparison. Build query groups, define page filters and compare the 28-day periods. Identify changes in impressions, position and CTR.
- Week 3: Priority updates. Select the 5 exposed guides with the most impressions, considering business relevance. Add a useful reason to visit and a clear next step. Log changes and publication dates.
- Week 4: KPI report and decision notes. Verify event recording. Document why each page changed, what you will measure and when you will review it.
Keep query, country, device and page scope consistent after updates. Note demand changes. If sufficient data are unavailable by the final week, defer conclusions until an appropriate observation window has elapsed.
Sources
These sources support the research findings and Google’s reporting guidance. Content grouping, update priorities and the implementation schedule are DijitalPi’s recommendations. Read the paper as a preprint and the platform documentation as reporting guidance; neither directly establishes the cause of traffic changes on your website.
- AI in Search Reduces Publisher Referrals Without Improving User Experience: Experimental Evidence, arXiv preprint
- Google Search Central: AI features and your website, updated 10 December 2025
- Search Console Help: Performance report advanced filtering
- DijitalPi research assessment: Does Google AI Mode reduce clicks to websites?



