Amazon says Alexa+ customers engage 2 to 3 times more and use integrated services 6 times more from discovery to booking. When the assistant runs more of the journey, your catalogue and content readiness decide whether you make the shortlist. Here is what the numbers mean and how to prepare.
Alexa Is Now a Transaction Surface, Not a Voice Toy: What Amazon’s Engagement Jump Means for Brands
By Stephen Honight, Founder of Lmo7
For most of a decade, Alexa was the smart speaker that set timers and played the radio. In July, Amazon published a set of engagement numbers that describe something quite different, and I think most consumer brands have not clocked what they imply.
The figures come from Amazon’s own Alexa+ developer announcement on 23 July 2026, so treat them as vendor launch numbers rather than independent data. I will come back to that caveat properly. Customers on the upgraded assistant are engaging with Alexa 2 to 3 times more than before, across a broader range of topics and tasks. And from initial discovery to final booking, they are engaging with services integrated into Alexa+ 6 times more. Amazon’s examples are not “what’s the weather”. They are buying event tickets, completing restaurant reservations and booking services without leaving the conversation. Modern Retail has separately reported Amazon saying customers make three times more purchases using Alexa+ on devices than they did on the original version.
The practical question underneath all of it is simple. For twenty years the job was “do we rank”. When an assistant builds the answer, the job becomes “does the assistant recall us”. Most of this piece is about what that change means and what to do about it.
The same announcement carried two structural changes that matter more than the stats. Amazon Wallet now lets a customer check out inside the conversation using the payment method already saved to their Amazon account. And Amazon has adopted the Model Context Protocol, so service brands can plug their own systems into Alexa+ directly. Priceline, Atom Tickets, Fandango and Taskrabbit are among the first names building task completion on it, per Amazon’s own list.
Put those together and the label changes. This is not a voice interface with a shopping feature bolted on. It is a transaction surface with a conversation on the front.
Read the numbers like an operator
Before anyone reorganises a budget around this, the honest caveats. These are Amazon’s own figures about Amazon’s own assistant, published in launch communications aimed at developers. There are no baselines attached, no absolute volumes and no category splits. A 6x lift on a small base can still be a small number. I would treat every figure here as directional rather than decision-grade, the same standard we hold any vendor study to.
But the direction is the story. Amazon does not publish engagement multiples to be modest, and it does not build a payments layer into an assistant that nobody transacts through. The company is telling brands, in its own numbers and its own product decisions, that the assistant is moving from answering questions to completing tasks. That is the shift worth planning for.
What changes when the assistant runs the journey
The mechanism matters more than the stats, so here is the way I would explain it.
A traditional Amazon journey has stages you can see and influence separately. The shopper searches, scans a results page, opens two or three listings, reads reviews, maybe leaves and comes back, then buys. Every stage is a surface: a ranking to win, an image to improve, a price to sharpen. Brands have twenty years of muscle memory for this.
An assistant-led journey compresses those stages into one conversation. The shopper describes a need. The assistant interprets it, recalls a shortlist, answers the follow-up questions and increasingly completes the transaction too. Amazon’s “discovery through booking” framing is precise: the journey starts and finishes inside the assistant. With Amazon Wallet in the loop, even the checkout stops being a separate place you can win someone back.
That compression does two commercial things. It shortens the shortlist, because a spoken or conversational answer carries two or three options rather than a page of twenty. And it removes the recovery points. On a results page, position eight still gets seen by somebody. In an assistant answer, the third option is close to invisible and the eighth does not exist.
Which is why the question changes from ranking to recall. Getting recalled when the assistant builds the shortlist is a different problem from ranking on a results page, with different work behind it.
The shopping side reads your catalogue
For consumer product brands, the shopping side of this assistant is Alexa for Shopping, the evolution of what Amazon launched as Rufus. And the thing to internalise is that it builds its answers from your Amazon catalogue: the product detail page (the PDP), your titles and bullets, A+ content, product attributes, the Q&A section and the review base.
That is a different retrieval problem from web search. There is no crawler to please and no backlink profile to build. The assistant already has full access to everything on your listing. The constraint is not whether it can read your content but whether your content contains anything worth recalling when a shopper asks a buying question.
Two patterns from our live client work are worth carrying into your planning.
The assistant asks category questions, not product questions. In our Rufus tracking across client accounts, the questions surfacing on listings are things like “which electrolyte product suits long rides” rather than “tell me about this SKU”. If your content only describes the product and never answers the category question, you are giving the model nothing to recall you with.
And content is read after eligibility is checked. The behavioural signals, meaning reviews, ratings, Buy Box health and sales rank, gate which products the assistant considers seriously before your copy gets its turn. I wrote about this at length in the Publicis Decoding Rufus piece. The practical read is not “content does not matter”. It is that content and eligibility are a pair: eligibility gets you considered and content wins you the recommendation. Run both tracks together.
We see the same pairing in the work. On Veloforte’s Amazon catalogue we have rewritten titles and A+ content across the range as part of a structured optimisation programme, with 42 improved titles live and 35 A+ content pieces deployed, precisely because the listing is what the assistant reads. And with Haleon we track a weekly prompt set across Alexa for Shopping, ChatGPT and Gemini on the Centrum business, because you cannot manage recall you are not measuring.
What “6x more, discovery through booking” implies
If Amazon’s engagement direction holds, a few practical implications follow for consumer brands.
More buying decisions get made without a browse. Every question the assistant answers in-conversation is a results page that never got rendered. The categories most exposed first are the ones where the question is easy to ask out loud and the risk of a wrong answer is low: replenishment, consumables, everyday health, pet, household. If a shopper can say “reorder something like the last one but cheaper”, the assistant can do the whole job.
Your content becomes your sales rep. In a compressed journey there is no second touchpoint where a human reconsiders. The assistant’s summary of your product, built from your listing, is the pitch. Vague bullets produce vague recommendations. Specific, honest, question-shaped content produces recommendations that hold up when the shopper asks a follow-up.
And the gap between prepared and unprepared brands widens quietly. Assistant recall does not show up in your traffic reports the way a ranking drop does. A brand can lose the shortlist months before it notices the sales effect, because the shopper who never saw you leaves no trace in your analytics. That is an argument for tracking, not for panic.
The catalogue-readiness work
My recommendation would be to treat this as a bounded readiness programme rather than a vague transformation. The work splits into four jobs.
Make the product page answer category questions. Rewrite titles and bullets in plain language a model can quote: who the product suits, what it does well, what it does not do. Use the honest works-for and does-not-work-for framing. Keep the claims you are allowed to make on the page, because a claim that lives in your internal document is a claim the assistant cannot cite.
Seed the Q&A layer at category level. Take the questions the assistant is actually asking in your category and make sure the answers exist on your listings, in A+ content blocks or the Q&A section. This is live territory: in our client work the category-level question pattern shows up consistently, and listings that answer those questions give the model something to retrieve.
Close the structured-data gaps. Complete backend attributes, keep parent-child variation structure clean so reviews pool rather than fragment, and keep the basics of eligibility healthy: review velocity, rating floor, Buy Box position, stock. This is the enabler track that makes the content work pay.
Track whether the assistant surfaces you. Build a prompt set for your category, run it on a weekly cadence and read the trend. The Haleon Centrum programme runs 184 weekly prompts across Alexa for Shopping, ChatGPT and Gemini for exactly this reason. Recall is measurable, and measured recall is the difference between reacting to a sales dip and seeing the shift coming.
The honest limits
Three things I would not want anyone to take from this piece.
Assistant-led purchasing is still the smaller share of Amazon volume. The browse journey is not dead and your ranking, pricing and advertising work still carries most of today’s revenue. This is about where the growth in influence is, not where all the volume sits today.
Amazon’s figures are Amazon’s figures. Until third-party panels size the assistant journey independently, hold the multiples loosely and plan on the direction rather than the decimal.
And content readiness alone will not carry a weak account. If your reviews, rating or Buy Box health are below the bar, fix that alongside the content, not after it. The two tracks run together, and pretending one substitutes for the other is how brands end up disappointed.
What to do next
If Alexa has been sitting in the “later” pile, the sequence I would run:
Get the recall baseline. Before spending anything, find out whether the assistant surfaces you today: run a category prompt set against Alexa for Shopping and the web assistants, and pull the data behind your eligibility signals. Our DaaS tier gives your team direct API access to run this data layer yourselves, from £250 + VAT per month per marketplace, with onboarding so you can read it.
Fund the catalogue rewrite where the baseline says you are weak. Question-shaped PDP and A+ content, category Q&A seeding and clean variation structure. That is the Amazon Content Optimisation module in our Challenger tier at £750 per month, and the Agentic Tracking & Content Optimisation module at £1,000 per month adds the weekly assistant tracking so every rewrite is tested against actual recall.
For multi-brand portfolios, prioritise by exposure. Replenishment-heavy and question-led categories feel this first, so the portfolio question is which brands need assistant readiness now and which can wait a year. That is an Enterprise conversation: multi-brand tracking and prioritised recommendations, from £5,000.
Amazon has told the market, in its own numbers, that the assistant now runs more of the journey. The brands that treat that as a catalogue-readiness job this year will be the ones the assistant recalls next year.
Stephen Honight is the Founder of Lmo7, the AI-native agency helping consumer brands win in AI-powered discovery and agentic commerce. Lmo7 works with brands including Trip Drinks, Veloforte, Brown-Forman, Haleon, Pelotan and Symprove across Amazon, D2C and AI search channels.