Voice Search in 2026: How AI Assistants Pick Their Sources

By Sean Chaudhary · Published Oct 16, 2026

We first wrote about voice search in 2022, when the question was whether “Hey Siri” and “OK Google” would change how people shop. Four years later, voice is back in a bigger form. People now talk to ChatGPT, Gemini, Siri, Alexa+ and Copilot the way they talk to a knowledgeable friend, and those assistants answer by reading the web in real time.

So “voice SEO” now means one thing: being the source an assistant reads, trusts and quotes when someone asks out loud.

This guide covers what changed since 2022, how AI assistants choose their sources, what ecommerce and B2B teams should do about it, and a checklist you can work through this quarter.

What changed since 2022

In 2022, voice assistants were mostly command tools. They set timers, played music and read a single featured snippet aloud. Shopping by voice was a small niche, and most spoken questions ended in a short answer with no source named.

Today’s assistants hold real conversations, and the audience is huge:

• ChatGPT reached 900 million weekly active users in February 2026, and its voice mode lets people talk through a purchase or a vendor decision while their hands are busy.
• Google says the Gemini app has 950 million monthly active users, and AI Mode in Search has passed 1 billion monthly active users (Alphabet Q2 2026 earnings remarks).
• Google also reported that Gemini Live conversations are five times longer than text-based conversations on average. Spoken sessions go deeper, with more follow-up questions.
• Amazon said more than 350 million shoppers used Alexa for Shopping over the 12 months to its Q2 2026 results, as Alexa+ rolled out more widely.
• Siri with Apple Intelligence has been able to pass requests to ChatGPT since iOS 18.2. In June 2026 Apple announced a more conversational Siri built on Apple Foundation Models developed in collaboration with Google’s Gemini models, with a user beta planned for later this year.

The traffic is starting to show up for retailers. Adobe found that traffic to US retail sites from generative AI tools rose 693.4% year over year in the 2025 holiday season. The starting base was small, but the direction is clear.

B2B buyers are moving too. In a Responsive survey of 350 B2B buyers (October 2025), a quarter said they now use generative AI more than traditional search when researching vendors, and nearly two-thirds use it at least as much as search.

What we see in our own data

We wanted to know how much of this reaches a mid-size B2B site like ours, so we checked our own numbers.

In 30 days of our Cloudflare logs, AI assistants fetched our pages live for users 2,163 times. Those are requests from agents such as ChatGPT-User and Claude-User, which visit a page because a person asked a question right then. Each one is a real conversation where an assistant decided our page was worth reading. We break down the full method in our log file analysis for AI search.

We also ran our AI query classifier on our own Search Console data. Of 10,486 queries, 311 (3%) were AI-assistant-style queries: long, conversational and phrased like a request to a person. That share is small today, and it is the part of search that looks most like spoken questions.

Both numbers are invisible in a standard analytics report. Most assistant visits never run your analytics script, so you need server logs or edge logs to see them.

How AI assistants pick their sources

When someone asks an assistant a question by voice, the assistant usually runs one or more web searches, fetches a handful of pages, and builds a spoken answer from them. If you want to understand that pipeline in more detail, start with our explainer on how AI search works. Five factors decide whether your page makes the cut.

1. Crawl access

An assistant can only quote what it can fetch. OpenAI, for example, documents separate agents: OAI-SearchBot surfaces sites in ChatGPT search, while ChatGPT-User visits pages when a user asks a question. Block the wrong one in robots.txt or in a firewall rule and you drop out of answers without any warning.

Check robots.txt, your CDN bot settings and any security plugin. Also check that key content is in the HTML. Many assistant fetchers do not run JavaScript, so content that only appears after scripts load may never be seen.

2. Clear, quotable answers

Assistants favor pages that answer a question directly near the top, in plain language, and then add detail. A spoken answer is short, so the assistant looks for a sentence or two it can lift with confidence.

Write the answer first, then the explanation. Use question-style headings that match how people speak. Keep each section focused on one question.

3. Structured data

Schema markup gives assistants facts in a format they can read without guessing. Product, Offer, Review, FAQPage, Organization and LocalBusiness schema all help. The rule is simple: the markup must match what is visible on the page. Mismatched schema undermines trust.

4. Entity clarity

Assistants need to be sure who you are. Use the same business name, description, address and category everywhere: your site, Google Business Profile, LinkedIn, directories and review sites. Add an About page that states what you do, for whom and where. Link your social profiles from Organization schema with sameAs.

5. Reviews and local data

For “near me” questions and product recommendations, assistants lean on review platforms, maps data and business profiles. The 2022 advice to claim your Google Business Profile still holds, and now it extends to Apple Business Connect, Bing Places, Yelp and the review sites in your category. Recent, detailed reviews give an assistant something specific to say about you.

What ecommerce teams should do

Shopping questions are where voice assistants are moving fastest. Google says its Shopping Graph holds more than 50 billion product listings, with more than 2 billion refreshed every hour, and AI Mode draws on it for conversational shopping. Your product data is your voice SEO.

Treat product feeds as a primary channel

Keep Google Merchant Center feeds complete and accurate: titles, GTINs, brand, price, availability, shipping and return details. OpenAI has published a product feed specification for ChatGPT that asks merchants for identifiers, descriptions, pricing, inventory, media and fulfillment options, and notes that recommended attributes like reviews improve ranking and relevance. If you sell on Amazon, the same discipline applies to your listings for Alexa+.

Write product pages for spoken questions

People ask assistants things like “what’s a good espresso machine for a small kitchen under $500” or “which running shoes are best for flat feet.” Answer those use cases on the page. Include dimensions, materials, compatibility, who the product suits and who it does not suit. Vague marketing copy gives an assistant nothing to repeat.

Build real FAQs from real questions

Pull questions from customer service tickets, live chat logs, reviews and Search Console. Answer each in two or three sentences on the relevant product or category page, then mark them up with FAQPage schema that matches the visible text.

Make comparison and category pages useful

Many voice shopping sessions start broad (“best gifts for a coffee lover”) and narrow down. Category pages with short buying guides, clear filters and honest comparisons give assistants a page to cite at the early stage, before the shopper picks a product.

What B2B teams should do

B2B buyers ask assistants to shortlist vendors, explain pricing models and compare options. Publish the pages those answers need: pricing or cost explainers, comparison pages, implementation timelines, industry pages and case studies with specific results. Name the industries, company sizes and locations you serve so an assistant can match you to the question. Our AI SEO services start with exactly this gap analysis.

The 2022 ideas that still work, updated

• Long-tail conversational queries. Still the core idea. The queries are now longer and arrive as follow-ups in a conversation, so cover the second and third question a buyer would ask.
• Position zero. Featured snippets matter less on their own. The same skill, a clear answer near the top of the page, is what gets you quoted by assistants.
• Page speed. Still important. Assistant fetchers work on tight time limits, and a slow or script-heavy page can be skipped.
• Local presence. Even more important. Assistants answer “near me” questions from business profiles, maps and reviews, so consistency across every listing counts.
• Natural language. Write the way your customers talk. Use their words for problems and products, along with your internal terms.

A practical checklist

Is voice search the future of ecommerce and SEO?

Voice search is now part of AI search. The assistants people talk to read the web, pick a few sources and speak an answer. The brands that show up are the ones that are easy to crawl, easy to understand and easy to quote.

The work is familiar SEO, sharpened for a new reader: accurate product data, clear answers, consistent entity information and strong reviews. Start with the checklist above and measure progress in your logs.

Want to know whether AI assistants are quoting you or your competitors? AlchemyLeads helps ecommerce and B2B brands get found and cited in ChatGPT, Gemini, Siri and AI Mode. See our AI SEO services and book a call with our team.