Google can win while a business loses the visit
Findability means a buyer can discover a business and act on what they find. Search is too large to write off. Alphabet reported $63.271 billion in Google Search and other revenue for the quarter ended 30 June 2026, up 17% from the same quarter in 2025. Revenue does not count searches or unpaid clicks, but it does show a company with the resources and incentive to change its interface rather than surrender it. Google has already put AI answers inside its own search product. [1]
That gives us a clear starting position: through 2029, plan as if Google remains a major route to customers. We would change that view if representative evidence showed valuable buying journeys leaving Google for sustained periods, along with a material loss of commercial referrals. A spike in chatbot accounts would not settle it. Nor would a single publisher's traffic chart.
Thailand makes the distinction sharper. Statcounter's September 2026 chart puts Google at 99.37% of the search engine referrals it measured to participating Thai websites. It does not say that 99.37% of Thai people use Google, or that Google starts 99.37% of buying decisions. The chart misses searches that produce no click and activity inside other apps; September data can still be revised under Statcounter's 45 day rule. It is strong evidence for keeping Google search in the plan and weak evidence for ignoring everything around it. Search, Maps and YouTube offer different interfaces under one owner, so appearing on all three does not diversify platform dependence. [2][3]
Search revenue can grow while a supplier loses informational visits. Revenue, referrals, answer mentions and booked work have different denominators.
The evidence points to fewer clicks on some answers
Pew tracked browsing by 900 US adults in March 2025 across 68,879 Google searches. People clicked a traditional result on 8% of visits to pages with an AI summary, against 15% without one. Summary citations drew a click on 1% of visits with a summary. These are shares of search page visits, not of people or all website traffic. [4]
Pew reconstructed the result pages in April, after the browsing, and longer questions were more likely to generate summaries. A factual query differs from a supplier search. The study observes behaviour without isolating the effect of the summary itself.
A later launch gives a second angle. Ahrefs used Google Search Console data from 963 high traffic French domains around the 22 July 2026 arrival of AI Overviews. Among domains where more than 20% of French queries triggered an Overview, median click through rate fell 23.1% relative to the prior 28 days during the nine days after launch. A group with less than 2% exposure rose 2.3%. That timing supports a loss of clicks for exposed domains, although nine summer days and a sample of large sites cannot tell a Thai service business what its own rate will be. The figure is a relative change in rate, not a 23 point drop in traffic. [5]
Google offers a serious counterpoint. In August 2025 it said total organic clicks from Search to websites had been relatively stable year over year and that clicks without a quick return had risen slightly. Google did not publish the counts, query mix or distribution across sites, so outsiders cannot reproduce that assertion. Still, it is not logically at odds with Pew: if people make more searches, fewer clicks per search can coexist with a stable total. A stable total can also conceal winners and losers. [6]
We therefore expect pressure on pages whose whole purpose is to provide a short, common answer. We do not predict that every website will receive fewer visits. A buyer seeking availability, a quote, a credible case, a room policy or a working booking route still has a reason to go beyond a summary. The site has to earn that visit.
A shortlist can form before the first site visit
More people now use AI for information, but usage and substitution are different claims. In Pew's February 2026 survey of 5,119 US adults, 42% said they had ever used a chatbot to search for information, while 24% said they used a chatbot daily for any purpose. These are people reporting use, not the share of searches or sales moved out of Google. [7]
Referral estimates show the same distinction. Similarweb put AI platform referrals at an average of 770.7 million worldwide visits a month from June 2025 through May 2026, up 117.4% on the preceding year. It also described AI referrals as a low single digit share of total traffic for most sites. That is a growing route to websites from a smaller base, measured by a vendor panel rather than each site's own logs. It does not establish how often someone researches in AI, later searches a brand in Google, and finally contacts the business. [8]
The journey already extends beyond general web search. LINE MAN reported more than 16 million searches for three food categories within its platform during its 2025 study period. That number is searches inside one service, not a share of Thai consumers, but it proves the practical point: a food choice can begin where the customer already orders. LINE Thailand separately reported more than 12 billion Official Account conversations in 2025. Those conversations are not unique people or sales; they show that chat is a working business interface at substantial scale. [9][10]
Imagine a Bangkok restaurant that appears in a friend's video. A guest checks its location and reviews, then asks its LINE account whether a wheelchair can reach a table. A polished site with beautiful photos cannot rescue an unanswered message or a missing accessibility fact at that moment. This is a constructed example, not a claim about typical Thai behaviour. It shows why findability is more than being mentioned: the person must be able to verify the choice and take the next step.
A firm buying a service may hear a peer recommendation, compare options with an assistant, then use Google to check the provider's proof and terms. We would ask recent buyers where their shortlist came from before paying to appear on another platform.
Three plausible paths from 2027 to 2029
These are planning scenarios, not shares of a probability forecast. They can coexist across categories. Our base case is that Google absorbs much of the interface change. We will still prepare for a more divided journey, and we will watch delegated purchasing where it actually works. None of these requires Google to disappear.
Google absorbs AI into everyday search
People continue to start many decisions with Google, but the result page answers more simple questions itself. Maps, video, reviews, ordinary links and AI responses sit in one familiar place. Businesses still get visitors for comparison, proof and action, even if fewer people open a page to read a definition. We treat this as the base planning case because Google has current distribution and resources, not because the 2029 outcome is known.
What would make this happen?
AI answers must remain useful inside Search while buyers continue to trust Google for commercial questions. The case weakens if a representative longitudinal panel shows sustained movement of valuable journey starts and customer acquisition to other services, beyond a rise in chatbot registrations.
Discovery divides by the job
A customer uses one place for inspiration, another for comparison and another to buy or book. Search remains large but stops being a reliable proxy for every first encounter. A restaurant, hotel or supplier may need fewer, better maintained surfaces that serve distinct buyer questions, rather than a presence on every app.
What would make this happen?
Customers must keep using different tools for different stages, with measurable shortlists or orders influenced outside general search. Repeated interviews and same customer journey data showing Google still starts and shapes nearly all relevant choices would weaken this case.
Agents handle repeat, low risk purchases
A person gives a budget, timing and preference, then lets an intermediary narrow options and complete a familiar purchase. We see this as conditional and more plausible first for repeat goods or simple bookings than for a costly service with legal or operational risk. Suppliers still need accurate inventory, terms, payment handling and support, even if the customer never opens the homepage.
What would make this happen?
Agents must complete transactions reliably, merchants must accept the cost and control tradeoffs, and buyers must use them repeatedly. This case weakens if completed delegated orders stay marginal or people routinely return to merchant sites before committing. A product demo alone is not adoption.
The limits are visible in commerce today. OpenAI launched chat checkout in 2025, then said in March 2026 that its initial version lacked the flexibility it wanted. Its newer shopping experience focuses on discovery. Buyers can complete purchases on Shopify merchants' stores through an in app browser, while a Walmart app inside ChatGPT supports Walmart payments. Neither route comes with a public share of completed agent orders. Capability has arrived; broad transaction habit has not been demonstrated by that announcement. [11]
Build for a decision, then measure the handoff
For DLVX, the operational answer starts with the buyer's next question. On the site, make claims verifiable: show what is offered, where it is available, what it costs when a price can be stated, how the work is done, and what happens after contact. For a local business, check hours, location, access details and the actual reply path. For a complex service, give a useful example, clear scope and evidence a colleague can share internally. A page that merely repeats the answer a search engine can provide has a weak reason to receive a visit.
Choose the channels customers actually use, keep essential facts current and make the next action work. An open restaurant with a broken booking link still has a findability problem. Accurate facts help, but no markup guarantees a recommendation. Google says ordinary search fundamentals still apply and special AI markup is not required. [12]
Measurement has to reflect those steps. Keep Search Console clicks, impressions and click through rate by query intent, language, device and country. Watch branded searches and direct visits for signs that a buyer learned the name elsewhere. Ask new customers what first put the business on their list, then record qualified enquiries, accepted quotes, bookings, repeat orders and margin. A direct AI referrer is useful evidence of one route, but it will miss an assistant conversation followed by a typed brand search. Do not claim every rise in branded search came from AI.
We would review the mix quarterly, starting with a small set of customer journeys rather than a universal dashboard. If comparable informational queries keep their external click rate as AI exposure rises, we should soften the click loss thesis. If customers routinely reach a decision and transact inside agents without visiting suppliers, we should change the site's job and the channel mix. And if Google continues to originate most valuable leads, the sensible response is to keep earning that demand.
By 2029, we expect Google and websites still to matter. What becomes less reliable is the assumption that being found, being clicked and being chosen happen in the same place. A business that helps a buyer verify and act will matter even when the first answer appears somewhere else.
Sources and methods
Published 20 October 2026. Study periods and publication dates differ; each source below states its scope. Forecasts are DLVX editorial judgements, not measured outcomes. Several AI models helped research and challenge this piece; their agreement is not evidence. The business examples are illustrative.
- Alphabet Q2 2026 results22 July 2026 SEC filing; quarter ended 30 June 2026; Search and other revenue, not clicks.
- Statcounter search engine market share in ThailandSeptember 2026 measured search referrals to instrumented sites; rolling page checked 20 October 2026.
- Statcounter methodologyUndated FAQ checked 20 October 2026; referral denominator and 45 day revisions.
- Pew Research Center, Google users and AI summaries22 July 2025; 900 US adults, 68,879 Google searches in March 2025; observational next visit analysis.
- Ahrefs, AI Overviews and French click through rates16 September 2026; 963 high traffic domains, 28 days before and nine days after French rollout.
- Google Search, AI in Search and click quality6 August 2025; first party aggregate click claim without published underlying series.
- Pew Research Center, Americans and AI 202617 June 2026; survey of 5,119 US adults in February 2026; self reported use.
- Similarweb, AI referral traffic by industry3 September 2026; modeled worldwide referral visits, June 2025 through May 2026.
- LINE MAN Wongnai, 2025 food trends12 December 2025; publisher calls the period 2025, although the release preceded year end and gives no exact extraction cutoff; company reported searches inside its food platform.
- LINE Thailand Business Insight 20265 August 2026; company reported Official Account conversations in 2025.
- OpenAI, Powering Product Discovery in ChatGPT24 March 2026; product announcement describing merchant store checkout path and commerce pivot.
- Google Search Central, AI optimization guideOfficial living guidance checked 20 October 2026; ordinary search fundamentals and no special AI markup.