When Google impressions remain healthy but website visits decline, it is tempting to blame the new answer at the top of the results. Understanding why AI Overviews cut clicks can help explain part of that behaviour, because some readers receive enough information without opening a website. It does not establish the cause of every decline, so the first task is to identify which searches and pages have actually changed.
Establish whether demand or click-through has fallen
Compare equivalent periods in Search Console, filtering for Malaysia and separating mobile from desktop where the difference matters. Review impressions, clicks and average position for the affected pages and queries. Falling impressions can indicate reduced demand or lost visibility, while stable impressions with fewer clicks point towards a different investigation.
Consider a hypothetical page receiving 10,000 impressions in each of two periods. If clicks fall from 500 to 300, its click-through rate has dropped from 5% to 3%. That calculation describes the change; it does not identify the cause. A different mix of queries, competing results or a less appealing title could also contribute.
Separate quick answers from purchase decisions
A definition page may satisfy a reader once the definition appears in the results. Someone seeking an installation quote or checking whether a particular item is in stock has a different reason to visit. Group pages by what the reader is trying to do, then assess whether the decline is concentrated in easily summarised information.
Keep brand searches separate from broader discovery searches too. A campaign ending can reduce searches for the business name, while seasonal demand can change the overall figures. For Malaysian sites, compare the relevant holiday or promotion period as well as calendar dates when those events materially affect demand.
Inspect the search results behind the pattern
Review a sample of the affected queries using appropriate Malaysian context and record what appears. Look beyond the Overview to shopping results, advertisements and competing pages. Save the date and query wording so later checks can distinguish a repeated pattern from a single observation.
Search Console’s Generative AI performance report can show impressions for supported AI features, with breakdowns including pages, countries and devices. It provides additional evidence of exposure, but an impression is not a click or sale. Read it alongside ordinary search performance and website analytics rather than treating it as a complete explanation of lost traffic.
Check technical changes before changing the content strategy
If a sharp fall affects many unrelated pages, review recent releases, redirects, indexing problems and measurement changes. A broken analytics tag can reduce recorded visits without changing search demand, while an indexing fault can remove pages from consideration altogether. Those problems need specific fixes, not a blanket rewrite of the blog.
Where the evidence points towards lower click-through on informational searches, consider what the reader still needs after receiving a short answer. A sizing calculator, documented comparison or detailed installation example can justify a visit when it is genuinely useful. The added material should help complete the task, rather than merely make the page longer.
Protect the value of the visits that remain
Review enquiries, sales and the quality of leads alongside traffic. Losing low-intent visits has a different commercial effect from losing enquiries about a profitable service. Improve the path from relevant advice to the next action, and check that forms, product selection and mobile purchasing work properly before spending more to replace the lost visits.
Respond to the pattern your own data supports and keep testing that explanation over time. Wider research provides context, but it should not become a forecast for every Malaysian site. Pew Research Center’s study of click behaviour when AI summaries appear found lower click rates in its US sample using March 2025 browsing data, which supports taking the issue seriously without assuming the same effect in Malaysia. A measured response combines diagnosis with more useful content and a stronger route from each relevant visit to a business result.