Need to Know News - September 23rd, 2026
In this week's Need to Know News edition:
🤖 The new Opus 5.5 brings near-flagship smarts at a much friendlier price...and it keeps its facts straight better than ever.
🤖 The Claude team published a hands-on playbook for Opus 5.5...one quick deletion from your prompts gets you faster answers.
🤖 Meta's AI agent can now check out at any Shopify store...and Shopify's CFO has a conversion stat that shows why it could be genuine needle mover for your store.
And a whole lot more!
Claude Opus 5.5 Lands Near Fable-Level Work, 40% Cheaper Than Opus 5
Claude's newest model, Opus 5.5, went live Tuesday, and it lands close to Anthropic's best work for less than the Opus it replaces. It also passed a research test its predecessors failed: write up a company's quarter from a copy of the web where the earnings release was hard to find, without inventing one figure or quote.
It managed 16 of 18... Fable 5.1 and Opus 5 never did.
📋 The Details: Anthropic puts it at roughly Fable 5.1's level on most work, and it runs about 40% cheaper than Opus 5. Subscribers also get higher five-hour usage limits on most paid plans, plus a rate-limit reset to bank for later.
🎯 Why You Need to Know: A made-up number in a research summary is the error that burns you with a client, and in this test only the new Opus kept its figures traceable. It also follows the writing rules you give it better, so a pasted style guide should hold.
⚡ Your Move: Rerun the last research brief where an AI invented a figure, feed Opus 5.5 the same sources, and check its numbers against the originals.
Delete "Think Carefully" From Your Prompts, Says Claude's Opus 5.5 Guide
Some familiar prompting habits now get in Opus 5.5's way, per Claude's official guide, starting with any line telling it to think carefully. Opus 5.5 already thinks before every reply and sets its own depth, and in the guide's testing, cutting that line got replies started sooner with no clear drop in quality.
📋 The Details: Two more tips from Addy Osmani's playbook suit marketing work. Asking for a design that "avoids a generic look" mostly swaps one default for another, so name the habits you're sick of, like cream backgrounds or pill-shaped buttons. For research, add "Mark anything you couldn't confirm, and say where you looked."
🎯 Why You Need to Know: It's also the first Opus with Fable-level biology and cybersecurity safeguards, and when one trips, Claude moves your chat to an older model... and since the check reads the whole conversation, a file you pasted earlier can set it off.
⚡ Your Move: Skim the guide's checklist and give Opus 5.5 one real job, like your next campaign brief, in one message that says what "done" looks like. Then judge it by how little editing it needs.
🚀 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
OpenAI Adds GPT-6 Sol and Luna at Half the API Price
OpenAI's GPT-6 family now comes in cheaper sizes. Flagship GPT-6 Astra launched earlier this month, and GPT-6 Sol and GPT-6 Luna now fill in below it, sharing Astra's training methods at 50% less through the API than their GPT-5.6 versions. On OpenAI's own factuality test, Sol also makes about half as many mistakes as its predecessor.
📋 The Details: Sol now costs $2 per million tokens in and $10 out, and Luna drops to $0.10 and $0.50. Both are live in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users. Mind that factuality stat's footnote, though... OpenAI tested on conversations where users had already flagged an error.
🎯 Why You Need to Know: On AutomationBench, a test of business workflows across 47 tools, Sol scored 33.2% at $0.27 a task, and OpenAI says that beats Opus 5 at max effort for 9% of the cost.
⚡ Your Move: If a tool you pay for runs on GPT-5.6 Sol, ask when it switches and whether your price drops too.
Taboola's New Realize ID Links What People Read to What They Buy
Following one buyer across the open web keeps getting harder, since a single purchase decision now winds through news sites, search, price comparisons, and AI chatbots. Taboola's new Realize ID claims it can follow that trail anyway, stitching scattered identifiers into one profile of a buyer who's still deciding.
📋 The Details: The signals come from publishers running Taboola's technology, NBC News and Yahoo among them, with a reach of over 600 million daily active users, so it sees scroll depth, time on site, ad clicks, and moves like starting a quote. Taboola claims up to a 2.4x lift in conversion efficiency, though it never says compared with what.
🎯 Why You Need to Know: CEO Adam Singolda argues AI has pushed buyers' research onto dozens of independent sites beyond search and social, which are exactly the pages where Taboola operates. Realize ID aims to reach people there after interest shows up and before they commit elsewhere.
📡 Watch For: Before-and-after numbers from any brand quoted in the launch, like Allianz or Ergo. That's how you'll learn whether 2.4x holds up outside a press release.
Meta's Muse Tops the App Store, and Amazon Has Locked It Out
Shoppers are handing more of their errands to AI agents, and right now the one they're downloading is Meta's Muse. Two weeks after launch it sits at No. 1 on Apple's App Store, one spot above ChatGPT, and Sensor Tower counted more than 2.5 million downloads in that span, within reach of ChatGPT's 3.1 million and far past Claude's 200,000.
📋 The Details: Amazon wants none of that on its own site, though. The company confirmed to CBS News that it has blocked Muse from shopping there, citing security and user experience concerns.
🎯 Why You Need to Know: In a Coresight survey of more than 1,000 consumers, 30% said they'd use AI to compare holiday prices, and as many would use it to hunt promo codes... so your price and your promo end up stacked against a competitor's in one answer.
📡 Watch For: Whether Muse still holds No. 1 on Black Friday. GlobalData's Neil Saunders says shoppers are "still a little bit nervous" about agents doing the buying, and a ranking that lasts through November says that's fading.
Shopify Is Wiring Every Store's Checkout Into Meta's Muse
Shopify is opening its whole merchant base to Meta's Muse, giving the agent a checkout lane into every Shopify store. Shopify CEO Tobias Lütke announced the partnership Monday on X, promising agentic checkout through Shop Pay, meaning the agent completes the purchase itself.
📋 The Details: Shop Pay is Shopify's one-tap checkout, and it passed $400 billion in lifetime sales volume in June, so this is hardly a side door. Neither announcement came with a launch date, though.
🎯 Why You Need to Know: Shopify CFO Jeff Hoffmeister said Sept. 10 that shoppers who start searching in an AI chatbot are 2.5x more likely to land right on a product page, with roughly an 80% conversion lift. Put checkout inside the agent and that product page does the selling alone.
⚡ Your Move: On Shopify? Confirm Shop Pay is on, then reread your 10 best-selling product pages as if each were the only page a buyer ever sees.
🧵 The Thread: Amazon blocked Muse over the risk of it capturing and storing customer credentials, per reports Sunday, and Shopify announced checkout at every store the next day.
IAB Tech Lab Writes the Rulebook for AI Agents Answering Ad RFPs
AI agents are handling more media planning and buying, and one stretch still has no shared rulebook: the back-and-forth between a buyer's request for proposal, or RFP, and a signed buy, still done by hand. On Tuesday, IAB Tech Lab's AAMP 3.0, its framework for ad-buying agents, added a spec for that stretch called OpenProposal.
📋 The Details: OpenProposal gives publishers one agent-readable format for describing ad products and answering a brief, so a buying agent can rank proposals from far more publishers than a planner could. It plugs into standards buyers already use, and there's a nice safety catch... an agent that resends an unanswered order gets a confirmation instead of a duplicate buy. Public comment runs until October 22.
🎯 Why You Need to Know: PMG's Mike Treon says "the reasoning travels with the proposal," so sellers can explain why a package fits the brief. If you sell ad space, your pitch now has to persuade the buyer's software as well as the buyer.
📡 Watch For: Early findings from IAB Australia's seller-side pilot, which tests proposals and approvals in a sandbox before any live money moves.
Comscore Finds Gemini and Claude Eating Into ChatGPT's Prompt Share
People are spreading their AI questions across more assistants than at the start of the year. Comscore's Q2 AI Intelligence Report, built from its consumer measurement panel, found ChatGPT's share of all AI prompts fell from 70% in January to 50% in June, even as its visitors grew from 59 million to 99 million over the year... the rest of the field grew faster.
📋 The Details: Gemini gained the most, climbing from 17% of prompt volume to 30%, while Claude jumped from 2% to 11%. Google's results page is tilting the same way, with AI Overviews on 39.4% of desktop searches this June against 25.8% last July.
🎯 Why You Need to Know: Half of all prompts now land outside ChatGPT, so an AI-visibility check that only runs there covers half the room. Brands are turning up in those answers more often, too, with Comscore counting fashion mentions up 123% and health and beauty up 150% between January and March.
⚡ Your Move: Ask Gemini and Claude the five questions your buyers ask before buying, note which brands each one names, and compare that with ChatGPT's list.
Kroger Sees AI Shopping Assistant Driving Larger Baskets
Kroger has put an AI assistant between its shoppers and the shelf, and its first read on what that does to a shopping trip is bigger baskets. That surprised chief digital officer Yael Cosset, who told Groceryshop 2026 in Las Vegas on Tuesday that he'd expected detailed instructions to narrow its picks and shrink orders.
📋 The Details: Shoppers can talk to it or snap a handwritten recipe, and it turns a dinner plan for a family with mixed dietary needs into recipes and a filled basket. Cosset's theory is that skipping the list-building frees people to consider more, which he says helps Kroger and its consumer packaged goods partners alike.
🎯 Why You Need to Know: Kroger plans to fold its existing personalization into the assistant, so suggestions will reflect each shopper's habits. If your brand is on Kroger shelves, the assistant already picks some of what fills those bigger baskets.
📡 Watch For: Kroger's assistant placing an order the shopper didn't build item by item. Cosset said the tool could eventually reach some degree of autonomous shopping, and then the brand choice sits with the agent.
New Grok 4.7 Holds Its Price, Outscores Fable 5.1 on Legal Work
SpaceXAI's newest Grok arrived Monday aimed at coding and knowledge work, and it costs exactly what Grok 4.6 did despite a bigger base model trained longer on problems that take hours to crack. The surprise is legal work, where the company reports 19.6% on the Harvey Legal Agent Benchmark against 6.7% for Anthropic's Fable 5.1.
📋 The Details: That price is $2 per million tokens going in and $6 coming out, and a fast variant doubles output speed for double the cost. It's already in Cursor, Grok Build, and the Grok API, and SpaceXAI says it got better at documents and presentations, scoring 1695 on the GDPval professional-work test to Fable 5.1's 1735.
🎯 Why You Need to Know: Grok 4.7 charges a fifth of Fable 5.1's input price and under an eighth of its output price while staying close on professional work. Coding is where it trails, with Fable 5.1 ahead on CursorBench 51.8% to 46.3%.
📡 Watch For: Grok 4.7 showing up inside Grok Bot, the agent SpaceXAI trained it to run within, putting the document gains in front of people who never touch an API.
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]
