Voice assistants now answer with generative AI, and spoken answers are short, so they usually name only the one or two brands the assistant is most confident about. Google is replacing Assistant with Gemini, Amazon has launched Alexa+, Apple has released Siri AI and ChatGPT Voice can search the web mid-conversation. This post covers what changed, how spoken answers differ from screen answers and what brands should do about it.
The voice assistants that matter now
Voice has moved from command-based assistants to the same models behind AI chat, which means voice optimization is now largely AEO with extra constraints.
| Assistant | What changed (per the company) | Where it runs |
|---|---|---|
| Gemini | Google is moving mobile users from Google Assistant to Gemini; Gemini for Home replaces Assistant on speakers and displays; cars with Google built-in are getting Gemini via software update | Android phones, Google Home and Nest devices, cars, watches, earbuds |
| Search Live | A voice conversation with AI Mode in the Google app, with links shown on screen; Google says it has expanded to everywhere AI Mode is available | Google app on phones |
| Alexa+ | Amazon's generative assistant; free for Prime members, $19.99 a month otherwise; available across the US and Canada | Echo devices, Alexa app, Alexa.com |
| Siri AI | Released as a beta with iOS 27 in September 2026; Apple says it uses information from across the web | iPhone, iPad, Mac, Apple Watch, Vision Pro |
| ChatGPT Voice | OpenAI says users can ask for a web search during a voice conversation, with visual results shown in the chat | ChatGPT apps |
Rollouts vary by device, language and country, so check what's live in your markets.
How spoken answers differ
Spoken answers differ from screen answers in four ways that matter for brands.
- Fewer names. Listening to a list of eight options is tedious, so spoken answers tend to name a few. If you're not among the most confident choices, you may not be said at all.
- No visible citations. Some voice experiences show links on a screen (Google says Search Live does), but on a speaker the user often hears only the answer. Brand recall depends on your name being said and understood.
- More action. Voice requests skew toward doing: book, order, call, navigate, reorder. The assistant needs an integration or reliable data to complete them.
- Context of the moment. People ask voice assistants while cooking, driving or walking. Answers lean on location, time and what's open now.
Write facts that survive being read aloud
Content that works in voice states one fact per sentence, in words that make sense without a table or a screen.
- Lead with the answer. "The Model X weighs 1.2 kilograms" is quotable; "Weighing in at an impressively light…" is not.
- Avoid meaning that lives in layout. A comparison table may not translate to speech. Add a sentence summary: "The Pro plan adds phone support and unlimited users."
- Spell out what symbols hide. Write "from 9 a.m. to 6 p.m., Monday to Friday" near your hours table, and prices with currency.
- Keep passages short and self-contained, so a single sentence can be spoken without context. See the answer-first writing framework.
- Answer the conversational question people speak, often longer and more casual than typed queries. See how people search differently in AI assistants.
Make your brand name speakable
Voice adds a problem screens don't have: the assistant has to recognize your name when spoken and say it back in a way people recognize.
- Test whether assistants understand your name when you say it, including common mispronunciations
- If your name is unusual, coined or stylized ("Xyrl," "b.ready"), make sure your about page and profiles explain it in words: "Xyrl, pronounced 'swirl'"
- Use one consistent written form everywhere, so spoken and written references resolve to the same entity
- For products with model numbers, make sure the spoken version ("the twenty-four hundred") maps to the right product in your content
See brand consistency and entity signals.
Local and action data
Many voice questions are local and immediate: "Is the pharmacy on Main Street open?", "Book a table for four tonight." These answers come from business listings and partner platforms more than from your website.
- Google: keep your Google Business Profile complete and current.
- Apple: manage locations and brand details in Apple Business, which Apple announced in 2026 as the replacement for Business Connect. See Apple Intelligence and Siri.
- Amazon: Alexa+ completes tasks through partners. Amazon has highlighted integrations with OpenTable, Uber, Ticketmaster and Thumbtack and announced Expedia, Yelp, Angi and Square for 2026. Being listed accurately on the relevant partner platforms is how service businesses get into those answers. See Amazon Rufus and Alexa+.
See AEO for local businesses for the full local playbook.
What about speakable schema?
Google's speakable structured data is still documented in beta, and it's narrow: it marks sections of news articles for text-to-speech on Google Assistant-enabled devices, for US users with Google Home devices set to English and English-language publishers. Google recommends marking key points of about 20 to 30 seconds rather than whole articles.
If you're a news publisher in that market, it's worth implementing. For most brands it isn't a lever, and it isn't a general voice ranking factor.
Example: a voice test for a home services company
This is an illustrative test plan, not observed results.
A plumbing company in Denver speaks these prompts on Gemini, Alexa+, Siri AI and ChatGPT Voice:
- "Who's a good emergency plumber near me?"
- "Is [company] open on Sundays?"
- "How much does it cost to fix a leaking water heater in Denver?"
- "Book a plumber for tomorrow morning."
What to record: whether the company is named, whether the name is pronounced so a listener would recognize it, whether hours and service area are right, what price range is stated and which source it came from, and whether the assistant offers to book, and through which platform.
Likely fixes this kind of test reveals: inconsistent hours across listings, a missing pricing guide page, no presence on the booking platform an assistant uses, and a brand name that speech recognition mishears.
Measuring voice visibility
Voice is hard to measure at scale because answers are spoken and personalized. Practical approaches:
- Run a small, fixed set of voice prompts monthly on real devices and transcribe the answers
- Track the typed equivalents of the same prompts in your regular tracking, since the same models answer both
- Watch for direct-response signals: calls, bookings and partner-platform orders that spike without matching web traffic
- Ask "How did you hear about us?" with an option for voice assistants
The bottom line
Voice assistants now run on the same generative models as AI chat, so the AEO fundamentals apply, but spoken answers are shorter, name fewer brands and lean toward action. Write facts that make sense when read aloud, make sure your brand name is recognized and said correctly, keep listings and partner platforms accurate for local and transactional requests, and test by speaking real prompts on each assistant in the markets you serve.
