The page that gets read aloud

What is voice search optimization in 2026, and what changed?

Voice search optimization is the practice of structuring content, technical performance and location data so that a spoken query to Siri, Alexa or Google Assistant returns your business as the answer. What changed is who else reads that structure. The same choices that let an assistant speak your page aloud now decide whether an AI answer engine cites you at all, which is why voice search optimization and generative engine optimization have stopped being two separate projects.

The practical consequence is a merged backlog. A page built for a smart speaker and a page built to be quoted inside an AI Overview want the same things: a direct answer near the top, facts that survive being lifted out of context, markup that declares what each block is, and a load time fast enough to make the shortlist. Teams still running voice as a niche channel and AI search as next year’s problem are paying twice for one piece of work.

Who is actually asking, and what are they asking for?

Voice is a mainstream input rather than an experiment. Around 20.5% of people worldwide actively use voice search,¹ 35% of Americans aged 12 and over own at least one smart speaker, about 101 million people,³ and 76% of voice searches are near me or local queries.⁴

The local share is the figure that should move your roadmap first. If roughly three in four spoken queries carry location intent, the highest return work is not conversational copywriting. It is making sure opening hours, service areas, phone numbers and physical addresses are marked up correctly, consistent across directories, and accurate on the day somebody asks. Assistants resolve those queries from structured data long before they reach your prose.

Frequency is narrower than ownership, and worth keeping in proportion. An earlier consumer index put daily voice assistant use at 28% of consumers in the US and UK,⁵ which was already a large habitual base before AI assistants moved into the same slot on the same devices. Nobody should plan a channel strategy on voice alone. Everybody should plan for the fact that spoken queries and typed AI prompts now compete for one answer rather than ten links.

What is generative engine optimization, and how does it differ from SEO?

Generative engine optimization is the practice of building content so that AI systems such as Google AI Overviews, ChatGPT and Perplexity select and cite it when they synthesize an answer, rather than merely ranking it in a list of links. Classic SEO competes for a position. Generative engine optimization competes for inclusion in a paragraph that may never display a position at all.

Industry tracking of Google results puts AI Overviews on approximately 48% of tracked queries, up around 58% year over year.⁶ The click consequence is the part worth internalizing.

On informational queries, click-through rate falls from 1.41% to 0.64% when an AI Overview appears.⁶ That is less than half the traffic for the same rank on the same query. The ranking did not get worse. The page stopped being the destination and became a source.

Three disciplines now sit on top of each other, and confusing them is what produces duplicated work.

They are layers of one problem rather than rival methods. Answer engine optimization is the formatting layer: ask the question the way a person asks it, then answer it in two or three sentences that stand alone. Voice search optimization adds the local and technical constraints. Generative engine optimization adds the authority layer on top. None of it replaces classic SEO, because backlinks, domain history and named expertise still gate which pages an engine trusts enough to quote. Treat generative engine optimization as additive work, not a migration.

Why voice search optimization and GEO are now one job

Both channels are fed by the same extract. The featured snippet is the clearest evidence for it, and the AI Overview is essentially a snippet with more sources and a citation row attached. Position zero work, written off by some teams as a legacy tactic, is now doing double duty.

The page a smart speaker reads aloud and the page an AI engine quotes are the same page. What differs is the delivery, not the build.

That is a design constraint rather than an observation. One well built asset can serve a spoken query, an AI Overview and a chat assistant at the same time, provided the answer sits near the top in plain language, the supporting detail is factual rather than adjectival, and the markup tells a parser which block answers which question. Producing three variants of the same content for three channels is the failure mode, not the strategy.

How do you build a page an assistant will read and an engine will cite?

Five moves carry most of the weight, and they can be applied to pages you already own.

• Lead with the answer. Put a 40 to 60 word direct answer immediately under a heading phrased as the question somebody would actually speak. Everything else on the page supports that block rather than delaying it.
• Raise fact density, not keyword density. Concrete numbers with named sources every 150 to 200 words give an engine something worth lifting. Adjectives give it nothing to quote and nothing to verify.
• Declare what the page is. FAQPage, HowTo, LocalBusiness, Product and Organization schema tell a parser what each block means. Voice assistants and answer engines both read the markup before they read the copy.
• Treat speed as a ranking lever, not hygiene. Voice search results load in about 4.6 seconds, roughly 52% faster than the average web page.⁸ Slow pages are filtered out of the candidate set before quality is ever assessed.
• Keep the local record clean. With near me intent dominating spoken queries, accuracy on hours, address, service area and phone number does more for voice than any rewrite of the body copy.

Speed deserves a line of its own, because it is the item most teams file under somebody else’s backlog. The same work that lifts assistant selection also lifts revenue on small screens, which is why it is worth sequencing alongside mobile conversion work rather than raising it as an isolated technical ticket.

How do you optimize a product page for voice and AI shopping?

Commerce is where this gets concrete, and where most published guidance stops. A spoken shopping query and an AI shopping prompt both ask for a decision rather than a list. The page has to supply the facts a machine needs to make that decision on the shopper’s behalf, in a form it can read without guessing.

Product schema is the floor: price, currency, availability, GTIN or SKU, condition, shipping and return terms, and an aggregate rating with a real review count behind it. An assistant that cannot resolve availability will not recommend the item at all. Above that floor, three things separate product pages that get quoted from product pages that get skipped.

1. Answer the buying questions in text. Sizing, compatibility, materials, delivery windows and return windows written as explicit question and answer pairs, not buried inside a tabbed panel, an image or a PDF.
2. Publish specifics that can be compared. Dimensions, weight, capacity and warranty length in plain units. Comparison is exactly what a generative engine is doing when it chooses between two products, and it can only compare what it can parse.
3. Keep price and availability honest in the markup. A stale price in structured data is worse than no markup, because it teaches the engine to distrust the whole feed.

None of this replaces the commercial fundamentals. It changes where they have to be legible. Where this sits inside a wider digital marketing program, sequence the schema and the answer blocks before commissioning new content, because most catalogues already rank for the terms that matter and simply cannot be extracted cleanly.

How do you measure AI search optimization when there is no SERP?

You stop measuring position and start measuring citation. AI engines do not publish a ranked list, so the reportable unit becomes the share of target prompts on which your domain appears as a cited source, tracked across the engines your buyers actually use rather than the one you find easiest to check.

Referral traffic is the second measure, and it is finally large enough to isolate. AI referred sessions grew 527% year over year in the first half of 2025,⁶ from a small base, but with a conversion profile worth segmenting in analytics rather than leaving buried inside direct traffic. Survey data also reports that 77% of US ChatGPT users treat it as a search engine,⁷ which is the habit change sitting underneath the referral numbers.

Four measures are enough to run this seriously:

• Citation share against a fixed list of 30 to 50 buyer prompts, re-run monthly on the same prompts so the comparison stays valid.
• Featured snippet ownership on that same question set, which remains the best available proxy for voice selection.
• AI referred sessions and their conversion rate, segmented out of direct traffic.
• Local pack and near me visibility for the queries that carry location intent.

Set the baseline before changing anything, because the honest version of this work is a comparison rather than a claim. If you want a read on which of your pages are already close to quotable and which need rebuilding, tell us the queries you want to own and we will show you where the gap actually sits.

Written by the group's editorial team with the practice leads who run these builds. Reviewed before publish. Spotted an error? Tell us and we will fix it.