The future of search: how AI is changing search and what your small business website needs now

The future of search is that people get an answer instead of a list of links: Google puts AI Overviews and AI Mode above the results, ChatGPT and Gemini become search destinations in their own right, and more and more questions come in through voice and camera. For a small business website that means fewer clicks per impression, but also a new opportunity: being named as a source in the answer. If you want to keep being found, you need a page that answers a question directly and verifiably, readable as text for every crawler.

Below: the numbers as of October 2026, the five shifts that matter, what specifically changes for voice and visual search, and what you do now as a business owner or marketer.

Where does search stand in October 2026?

AI search is no longer an experiment but the default experience for hundreds of millions of people. The numbers that matter:

• Google AI Mode has, according to Google, more than 1 billion monthly users since its global rollout in October 2025. Google has merged AI Overviews and AI Mode into one search experience.
• The Gemini app has 950 million monthly users, according to the same source.
• ChatGPT reached 900 million weekly users in February 2026.
• Americans use it. According to Pew Research Center, about a third of US adults had used ChatGPT by early 2025, and a clear majority of adults under 30.

Google is not disappearing: Alphabet’s search revenue grew 17 percent in the second quarter of 2026, and Google says it sends billions of clicks a week to websites through its AI features. Search is not moving away from Google; it is changing inside Google and next to it. Which AI platforms Americans use most is in what is an AI search engine.

The five shifts in search

1. From links to answers

The biggest change is that the searcher reads the answer before seeing a website, so fewer people click through per impression. Three measurements show it:

• Pew Research followed 900 US adults in March 2025: with an AI summary they clicked a regular result on 8 percent of visits, without a summary on 15 percent. A link in the summary itself was clicked on 1 percent of visits.
• Ahrefs measured in May 2026 across 300,000 keywords that an AI Overview lowers the click-through rate of position 1 by 58 percent.
• Semrush found that 92 to 94 percent of AI Mode sessions did not lead to a website visit (US desktop sessions, May to July 2025).

What that means for you: your website becomes less of a destination and more of a source. Whether you are named in the answer counts as much as your ranking. How AI Mode picks sources is in Google AI Overviews and AI Mode.

2. From keyword to question

Searchers no longer type two words; they ask a full question with context. “Which laptop suits me if I edit a lot of video, want to spend at most $1,500 and want a light model?” is one question in a conversation that can end in a purchase. A page written for a loose keyword does not answer that. A page with clear sections per sub-question does. How to plan for that is in content that AI cites.

3. From typing to talking

More and more questions are spoken and the answer is read aloud, which means there is only one answer and no second or third position. Google brought Search Live, searching with voice and camera inside AI Mode, to all countries and languages where AI Mode is available in March 2026. ChatGPT and Gemini have similar voice features. How to write for that is below under voice search.

4. From text to images

People search with their camera: a photo of a product, a plant or an error message, with the question added. That calls for images and video an AI model can understand. See multimodal search.

5. From searcher to agent

AI assistants are starting to search, compare and sometimes buy on behalf of the user. How such a system works is in what is an AI agent. Google says it is adding agents to its search experience, and OpenAI has introduced checkout inside ChatGPT. For an online store, structured product data becomes more important than the design of the category page. Read on in AI agents and the end of the website visit, ChatGPT Instant Checkout and agentic commerce for Shopify, WooCommerce and Magento.

On top of that there is privacy: AI assistants process personal questions and increasingly connect with email, calendars and documents. What that means is in AI search, privacy and data protection.

Voice search: how to become the answer that gets read aloud

For spoken questions, the page that gives a short, factual answer directly below a question, understandable without visuals, wins. Classic voice search turned a spoken question into a typed search and showed results. An AI assistant processes the question, writes an answer and speaks it. There is no results page to scroll through.

Spoken questions come through phones, smart speakers, cars and voice features in AI apps. They are longer and more natural than typed questions:

• Typed: “restaurant denver vegetarian”
• Spoken: “which restaurant in Denver has good vegetarian dishes and isn’t too expensive?”

What works for voice

• Use the question as a heading and answer it in the first sentence below.
• Keep the answer short. One or two sentences, short enough to read aloud and specific enough to say something. For example:

What does GreatCompany cost? GreatCompany Pro costs $49 a month for one website. Automatic for WordPress costs $99 a month.

• Write the way you talk. A natural tone fits a spoken answer better than jargon.
• Work with questions and answers. An FAQ block with the exact question as a heading and an answer of at most three sentences gives an assistant a ready-made answer. FAQ markup in structured data is fine, as long as it matches the visible text.
• Make local details complete. Many spoken questions are local: “plumber near me”, “is the pharmacy open now”. Keep your Google Business Profile current with hours, services and prices, and make sure name, address and phone number are the same everywhere. See GEO for local businesses.

What does not work for voice

• Long introductions before the answer comes.
• Information that only lives in a table or image; it cannot be read aloud.
• Text like “click here” or “see the image above”.

How do you measure it?

Not directly: Google Analytics has no separate voice segment. Look at signals: growth in traffic on long question queries in Search Console, more visits to your FAQ pages and more local searches. For AI assistants, you mainly measure whether you are mentioned.

Multimodal search: how to make images and video findable

Multimodal search means a question can consist of text, image, sound or video, and that an AI model only uses your images and videos well when there is text explaining what they show. ChatGPT, Gemini and Claude accept photos and screenshots; Google combines search with the camera through Lens and Search Live.

The scale is large. According to Google, Lens was used for almost 20 billion visual searches a month in October 2024, 20 percent of them with shopping intent. Circle to Search, which lets you circle something on your phone screen to search for it, was on more than 250 million devices in April 2025.

Examples of such questions: a photo of a plant disease with “what is this and how do I fix it?”, a screenshot of an error message with “how do I fix this?”, or a photo of a lamp with “where can I buy this?”.

Images

• A descriptive file name: not IMG_4523.jpg but a name that says what is in it.
• Alt text that describes what you see, not a row of keywords.
• Text around the image that explains what it shows and why it is there.
• Your own photos instead of stock photos used on thousands of sites.
• For online stores: several angles, the product in use, and images linked to your product feed with a GTIN or UPC.
• Structured data that links the image to the product or article. See the JSON-LD guide.

Video

• Put a full transcript on the page. An AI model reads the text; the video is the source.
• Add chapters with timestamps to longer videos, so a sub-question points to the right fragment.
• Use VideoObject structured data with a description, duration and thumbnail.
• Give YouTube titles and descriptions the question your audience asks.

The opportunity differs per sector. Online stores win with product recognition through the camera. Restaurants and travel win with good photos of dishes, rooms and locations, linked to structured data. Service providers with explainer videos only make them findable with a transcript.

What stays the same?

The foundation of being found does not change; only the layer above it does. Google writes in AI features and your website that there are no special requirements to appear in AI Overviews or AI Mode: your page must be indexed, internally linked, with the important content as text, and structured data that matches what is visible on the page.

• Quality wins. Google’s guideline Creating helpful content asks for original content that goes beyond the obvious, with clear sources and visible who wrote it and why.
• Authority counts. Sources that are reliable and consistent get named. See E-E-A-T for AI.
• Technical access stays the condition. GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot do not run JavaScript; only Googlebot renders. If your content only appears after scripts load, those crawlers see an empty page. See how AI crawlers read your site.

An llms.txt file is cheap and does no harm, but has no proven effect on citations: Ahrefs saw that 97 percent of llms.txt files on 137,210 domains received not a single request.

What does this mean for your small business website?

A small business website that wants to be found in AI search answers questions directly, shows who is behind it, is readable as text for every crawler and measures whether it gets mentioned. Concretely, in order of impact:

1. Put the answer at the top. Rankability saw in September 2026 that pages introducing the topic in the first 100 words were in the top 10 AI citations 22.1 percent of the time, against 8.7 percent for the rest. Pages covering the topic broadly scored 27.6 against 14.5 percent. It is correlation, not proven cause, but the direction is clear.
2. Write in questions and clear sections. A Semrush study of 304,805 URLs cited by language models found that cited pages scored clearly higher than non-cited pages on clarity and summarization (+32.8 percent), E-E-A-T signals (+30.6 percent), a question-and-answer format (+25.5 percent) and section structure (+22.9 percent).
3. Show who you are. A name, experience, real examples and sources. That is exactly what an AI model uses to trust a source.
4. Check that your content is in the HTML as text. The technical scan shows whether AI crawlers can read your pages.
5. Give images and video text. Alt text, captions and transcripts, as described above.
6. Keep your local details current. For spoken and local questions, your Google Business Profile is often the first source.
7. Measure whether you are named, not only whether people click. Clicks drop per impression; being named is the new ranking. A prompt tracker measures every day whether ChatGPT and other AI models mention your business. How to approach that organic channel is in what is GEO.

How far can you look ahead?

The shifts above are measured; what happens in 2027 is an expectation. Our predictions, and which of them held up by September 2026, are in 10 GEO trends for 2027. Want to know what changes for advertisers? Read advertising in AI search engines.

Read on

• Google AI Overviews and AI Mode
• What is an AI search engine?
• AI agents and the end of the website visit
• AI search, privacy and data protection
• 10 GEO trends for 2027
• Technical GEO checklist