Need to Know News - September 10th, 2026
In this week's Need to Know News edition:
🤖 Nearly a quarter of AI shoppers say a chatbot talked them out of a brand... and the reasons it gave are the boring ones.
🤖 OpenAI's image model now keeps your product intact while it rewrites the headline around it...
🤖 Time built an ad format for an audience that can't buy anything... and its COO thinks it's the most valuable placement on the site.
And a whole lot more!
23% of AI Shoppers Say AI Talked Them Out of a Brand
23% of people who shop with AI say it talked them out of a brand they were seriously considering. ICERTIAS surveyed 8,000 adults across 57 countries this summer, and the study explains how.
📋 The Details: 55% of respondents have shopped with AI, and 61% would rather have AI pick three products than compare 20 themselves... so fourth place now pays like not existing. 49% would buy an AI pick without visiting the maker's website, and shown an unfamiliar brand, 53% would likely buy on the AI's word alone, 62% with verified reviews, 65% with independent recognition.
🎯 Why You Need to Know: The reasons AI cuts a brand are dull ones, better value elsewhere or weaker specs, and when a preferred brand cost 10% more, only 24% would stay loyal.
âš¡ Your Move: Write down why your product costs more than the cheaper alternative, in specs a machine can read. If the answer is "people know us," the study says that holds a quarter of them.
Is Gradium An ElevenLabs Killer?
Giving your phone agent a Québécois receptionist's voice meant finding a real Québécoise and licensing her for the life of the product. Gradium's Voice Design skips the casting call: write the voice you want in a sentence and it generates one from the words alone.
📋 The Details: Type the brief in Gradium's Studio or API, get up to five candidates in seconds, and keep the one you like. It covers five languages, regional accents included. No human was recorded, so there's no actor license and no royalty. Gradium's own blind test, 7,627 comparisons, put it first in all five languages against ElevenLabs and three others, 97% on Québécois... the vendor ran the test, so treat the margins as marketing.
🎯 Why You Need to Know: A voice in the customer's own accent used to be reserved for the flagship product. At a sentence per voice, your Montreal line and your São Paulo line each get one.
âš¡ Your Move: Write the brief for your ideal support voice as if you were casting it, then play the candidates to your support team before anyone touches the phone tree.
🚀 WATCH: How These AI Copy Bots Are Producing World-Class Sales Copy 50X Faster Than Even The "BEST" Copywriters On The Market…
(Plus… They Don't Get Sick, Miss Deadlines, Or Ask For Raises Either!)
Watch the full AI Copywriting Tell-All Video Here
ChatGPT's Image Model Gets Sharper, Faster, and a Sketch Tool
More than 3 billion images come out of ChatGPT and its API every week, and on Tuesday the model behind them got a new version.
Images 2.5 renders sharper detail, keeps the people from your reference photos recognizable, follows edit instructions more precisely, and generates faster... and OpenAI wrapped it in fresh controls, from a Sketch tool for drawing your own reference to templates and on-image comments.
📋 The Details: Speed is what you'll feel first, since OpenAI says generation runs up to 50% faster than Images 2.0, and every ChatGPT, ChatGPT Work, and Codex user has it now. Ask for one change and the model leaves everything around it alone, even several turns into a conversation. Templates cover posters, flyers, product photos, and merch.
🎯 Why You Need to Know: Posters and product photos are Tuesday-afternoon work, and since an edit can now swap the headline copy without redrawing the product, you can spin a dozen versions off one approved image.
âš¡ Your Move: Run your best product image through it with three copy swaps and one background change, then check whether the product stayed put. That's the fidelity claim, tested on your own asset.
Meta's Muse Agent Shops With a Card Number the Merchant Never Sees
Send a WhatsApp message asking for a dentist appointment, and Meta's new agent goes and books it. Muse runs on its own virtual computer with a browser, so it fills out forms and keeps working after you've closed the app.
📋 The Details: You talk to it in the Muse app or straight inside WhatsApp, free up to a usage limit. It asks before it sends an email or buys anything, and every action lands in an audit trail.
The shopping part is the clever bit: at checkout Muse generates a one-time card number, so neither the merchant nor the agent sees your real card, and Meta says eligible purchases carry Link's purchase protections.
🎯 Why You Need to Know: A Muse order arrives at your store with a card number you've never seen and a browser session with no human behind it, so a legitimate purchase can look like the exact pattern a fraud filter is trained to block.
📡 Watch For: Pricing on the paid tier, and the 1Password integration Meta lists as coming soon. In your own order logs, checkouts failing on virtual card numbers.
Instacart Is Renting Its Shopping AI to Grocers Under Their Own Names
Grocers kept asking Instacart for its marketplace AI, minus the Instacart branding. On Tuesday they got it. Cart Assistant takes the technology behind Clementine, Instacart's new consumer shopping assistant, and drops it into a grocer's own website and app wearing the store's name, tone, and loyalty program.
📋 The Details: Food Bazaar, Heritage Grocers Group, and Woodman's are live, with ALDI U.S., Harmon's, Save Mart's banners, and Stew Leonard's lined up behind them. Shoppers get deals found, favorites reordered, and a handwritten list photographed into a cart, and it plugs in through Instacart's Storefront Pro platform. Heritage Grocers made a point of running it in Spanish.
🎯 Why You Need to Know: The assistant deciding what goes in the cart now lives on the grocer's own site, trained on that grocer's catalog and customer data... so if you're a brand on those shelves, the promotion you run with the retailer is what the assistant reads.
âš¡ Your Move: If you run a regional banner on Storefront Pro, ask your Instacart rep what the setup needs from you, since name, tone, loyalty hook, and recipes are yours to supply.
Gemini Will Now Write Your Brand Jingle, Vocals Included
Since last Thursday the Gemini app has had a music model that Google calls its best-sounding yet. Pick a genre, choose vocal or instrumental, and Lyria 3.5 hands back a song, with more expressive vocals and fuller arrangements than the last version.
📋 The Details: Everyone gets it globally on the web and in the mobile app, with templates that start you at background music or a custom birthday track, and a choice between short and longer pieces. Developers reach it through Google AI Studio, and it's also inside Google Vids and Flow Music.
🎯 Why You Need to Know: Google names the backing track for a video and the brand jingle as the intended uses, so the music bed for your next ad or explainer can come from the same app that drafts the script.
âš¡ Your Move: Make a 30-second instrumental bed in your brand's genre for the next social video and play it beside the stock track you'd normally pay for.
Contact Centers Rank the Handoff to a Human Dead Last
Ask 602 contact center leaders what they want from AI agents this year, and the top answer is keeping you out of the queue. Verint's State of Contact Center AI 2026 ranks containment first, ahead of revenue, customer experience, and cost... and it ranks getting the customer to a human when the bot fails ninth of nine.
📋 The Details: Almost 90% of these leaders plan to spend more on customer-facing AI this year, even though fewer than half say AI has significantly cut routine work for their agents. Separate research finds fewer than 60% of customers who've noticed AI in a service interaction think it made things better.
🎯 Why You Need to Know: Containment counts a conversation that ended and never asks whether the customer's problem did. If your bot's success number is the share it closed alone, the customers who gave up mid-chat look identical to the ones who got helped.
📡 Watch For: Whether handoff climbs Verint's ranking next year, and whether your own AI vendor can show a resolved rate beside its containment rate. If it can't, it's counting the way these 602 do.
Agencies Now Vet Creators by Whether Chatbots Cite Them
Creators used to answer to reach and engagement. Digiday reports a second judge now sits beside the audience: the AI models that cite creator videos when someone asks what to buy. At Tinuiti, AI visibility went from a tenth of a creator strategy to about a quarter in six to eight months.
📋 The Details: Long-form video is what gets cited, per Go Fish Digital's Lauren Lyster, who says Shorts rarely appear in AI answers. So Dept now hands creators exact product names to script from, builds chapter markers around the questions people ask, and insists on a clean transcript over auto-captions. Rate cards haven't moved... execs expect that in six to 12 months.
🎯 Why You Need to Know: A creator with 100,000 followers and real depth in one category can outweigh one with five million, since a model treats consistent expertise as evidence and keeps citing it after the campaign ends. Ogilvy's Charlie Coney adds a caveat: how models weigh sources isn't public, so there's guesswork here.
📡 Watch For: The first rate card that charges for citations separately from views. Jellyfish's John Dawson is already asking clients.
Time Magazine Is Selling Ads That Only Bots Will Read
When one of roughly 70 approved AI agents hits Time's website, it gets rerouted to a page no human sees: a plain text file carrying a sponsor's verified facts, labeled as an ad, before the bot gets through to the article. COO Mark Howard told Marketing Brew that bots outnumber human visitors on a growing number of days.
📋 The Details: The format is Agent Ads, built with ad-tech firm Mobian. For launch sponsor Ally Bank the page carries basics like its founding date, then a marker that the ad has ended. Howard's pitch rests on Ally's own data showing visitors arriving via AI referrals convert higher and land further down the funnel.
🎯 Why You Need to Know: Howard says agents store what they scrape and answer from memory later, so the facts a bot reads about you today are the facts it repeats for months, wrong ones included.
📡 Watch For: A second publisher signing with Mobian, and any conversion numbers Ally releases from the placement.
🧵 The Thread: Howard's "knowledge layer" is the shelf ICERTIAS found 49% of AI shoppers buy from without visiting your site.
ChatGPT Work Now Turns "Why Did Sales Slow?" Into a Dashboard
Somewhere in your company, the reason sales slowed last month is already sitting in a database, and getting it out means waiting on whoever can run the report. OpenAI's new Data agent in ChatGPT Work skips that line: it plugs into company data, works out what changed, and hands back a dashboard you can share.
📋 The Details: It connects to Snowflake, BigQuery, Databricks, and others, and it reads your company's own metric definitions, so "active customer" means what finance says it means. Every query respects the permissions the account already has, and admins install it as a plugin called Data and decide which roles get it.
🎯 Why You Need to Know: The customer quotes are OpenAI's hand-picked Alpha customers, so read them as a highlight reel, but CookUnity's habit is worth copying: its growth team asked for a peak-season conversion dashboard, then checked it against internal reports before spending on it.
âš¡ Your Move: On ChatGPT Work? Ask your admin to install Data, connect one source, and useOpenAI's own suggested prompt: diagnose why weekly active users changed last week, with likely drivers.
Thanks for reading.
Until next time!
The AI Marketers
P.S. Help shape the future of this newsletter – take a short 2-minute survey so we can deliver even better AI marketing insights, prompts, and tools.
[Take Survey Here]
