👋 Hey,
Welcome to the latest edition of The AI Driven Marketer!
Every week, I cut through the noise to bring you the most practical marketing AI news, tools, and a battle-tested automation workflow you can plug directly into your business.
Let's get into the build.
⚡ The AI Pulse: What Marketers Need to Know
Claude merges Chat and Cowork, and starts making documents and decks
Anthropic folded Claude Chat and Claude Cowork into a single interface on September 16, and brought Claude Design in with them. The company said people were tired of picking the right tool before starting a task, so Claude now works out what a request needs and pulls Cowork and Design capability into any conversation. Alongside the merge it launched Claude Docs and Claude Slides in beta: generate a document or a deck inside a chat, edit it by hand, and share it with collaborators who can leave comments. The rollout starts with Pro and Max users on web, desktop and mobile, with Team and Free to follow.
So What? The Chat versus Cowork decision that confused every non-technical user is gone, and so is a step in most content workflows. If your process was draft in Claude, paste into Google Docs, format, then share, the middle of that now becomes one shareable link.
A ChatGPT co-creator launched a model that cannot write a sentence
TypeSafe AI came out of stealth on September 15 with $40M in seed funding and Jev, which it calls a System One model, after the fast automatic thinking in Kahneman's work. Jev does not generate text at all. Software sends it a block of state plus typed questions, and it returns a choice, a score, or a yes-or-no probability, each carrying a calibrated confidence value, in 70 to 500 milliseconds. TypeSafe lists it at $0.042 per million input tokens with output free, and founder Diogo Almeida worked at OpenAI on the instruction-following research behind ChatGPT. Vercel called it the fastest-adopted model in AI Gateway history, reaching close to 13% of paid teams inside 24 hours, with Cloudflare, LangChain and Langfuse integrating within three days. The headline figures of 193.6x faster and 444.6x cheaper are TypeSafe's own, scored by agreement with GPT-6 Astra and Claude Fable 5.1 rather than against verified ground truth.
So What? Every classification, routing and scoring call in your stack currently runs on a model built to write prose that your code throws away. This is the first serious attempt to price those decisions separately, and the bonus below has six marketing jobs you can move across this week.
Canva bundles four professional apps and keeps the core ones free
Canva launched ProSuite on September 15, connecting Affinity for design and photo work, Cavalry for motion, Flourish for data visualisation and Leonardo for image generation, with more than 100 new features across them. The Affinity and Cavalry core apps stay free, with paid plans adding the AI features and team tools. Affinity picked up the long-requested Blend Tool, native 3D support, InDesign file import, a rebuilt RAW engine, and a scripting layer with a Claude connector that lets you describe an automation in plain language instead of writing it.
So What? A freelancer or small team can cover design, photo, motion and data visualisation without an Adobe subscription. The Claude connector is worth an afternoon on its own: batch jobs like resizing a campaign across fifteen formats or prepping print files become something you describe rather than build.
Google makes AI search visibility measurable, and India is in the first wave
On September 16 Google made AI performance insights generally available in Merchant Center across Australia, Canada, India, New Zealand and the US. The report compares your share of voice against other brands on AI Mode and AI Overviews. Google also opened a US beta putting its Business Agent inside YouTube ads so viewers can ask product questions without leaving the video, and added cart transfer and checkout testing to its Universal Commerce Protocol hub. Google says merchants who follow its core feed practices see roughly a 5% conversion increase the following month, and that in a Lululemon test its conversational attributes were used in about half of relevant AI Mode recommendations. Both figures come from Google and have not been independently audited.
So What? You can finally see whether AI answers mention you, and the lever is the product feed rather than the blog. Two fields cost nothing to fill in today: conversational attributes and loyalty data.
ChatGPT ads can now hold a conversation
OpenAI began testing Sponsored Agents on September 16. Someone who sees an ad in ChatGPT can open a separate, clearly labelled chat with an agent the advertiser sponsors, ask about sizing or compatibility or anything else that matters to them, then follow a link to the site when they are ready. OpenAI keeps that conversation walled off from ChatGPT's own answers and from the user's original thread. The same announcement added natural-language campaign building through an Ads Manager plugin, and made HubSpot its first CRM partner and Shopify its first ecommerce partner, both live the same day.
So What? The click stops being the moment of truth. When a sponsored agent fields buyer questions on your behalf, your product data is doing the selling, and thin or outdated copy loses the sale before anyone reaches your site. Testing is limited to selected US advertisers, so treat the gap as runway to fix the data.
Meta puts creator discovery and partnership ads in one place
Meta launched its Creator Marketing Hub globally on September 16 at IAB's Global Creator Week, merging Creator Marketplace and Partnership Ads Hub into one surface. Brands can filter for creators, find organic posts that already feature their products, turn a listing into an ad in one click, manage content permissions with expiry dates, strip copyrighted music before running a post as an ad, and message creators without leaving the hub. Facebook creators joined Instagram creators in the Creator Marketplace API, Meta's business assistant now handles conversational creator search inside the hub, and Partnership Ads are supported by the Meta Ads MCP connector. Instagram live video ads go generally available on September 29.
So What? Creator partnerships become a media-buying workflow instead of a week of DMs and file transfers. The MCP connector is the quiet part: you can run partnership ads from Claude or whichever agent you already work in.
🛠️ E-Books & Guides
🛠️ The AI Toolkit
3 Tools specifically curated for digital marketers.
1. Contentdrips
Best for: Solo creators and marketers who publish LinkedIn and Instagram carousels every week.
What it does: Turns a topic, blog post, YouTube video or TikTok into branded carousels and text posts written in your voice and designed in your brand kit, then publishes them straight to LinkedIn and Instagram, with an MCP server so Claude can run the whole pipeline.
Price: Free plan, then $15 or $26 a month.
2. DataHub Pro
Best for: Solo consultants and small agencies who build client reports out of spreadsheets.
What it does: Takes an Excel or CSV file and runs 67 ready-made analyses including cohort retention, RFM segmentation and Holt-Winters forecasting, then exports a branded Word or PowerPoint report, with every AI answer showing the operations it ran underneath.
Price: Free tier, then $19.99 a month for the workspace rather than per seat.
3. Prosperus
Best for: Freelancers and small studios who lose evenings to proposals.
What it does: Turns a project brief into a branded proposal with a timeline, package options and e-signature, then converts it into an invoice the moment the client signs.
Price: $15 a month, $12 a month billed annually, or $249 once.
Bonus
Six marketing jobs you can move off your expensive model
A model called Jev launched last week, and the part worth your attention is the shape of it rather than the price tag. It cannot write. You hand it a block of text plus a list of questions whose answers are defined in advance, and it hands back a pick, a score, or a probability, each with a confidence number attached. Roughly a tenth of a second per call. $0.042 per million input tokens, with output free, because there is no prose to meter.
That sounds narrow until you count how much of your AI spend already goes on exactly that. Sorting. Scoring. Deciding whether something deserves a human. Most of those calls today run on a model that writes a paragraph explaining itself, and your code throws the paragraph away.
Here are six marketing jobs where the decision is the entire job. The arithmetic below is list-price maths at stated volumes, published by people testing the model in the first week. None of it is an invoice, including mine.
1. Lead scoring and routing
The decision: does this lead match our ICP, how urgent is it, and who should call them. BetOnAI priced 5,000 leads a month at around 700 tokens per decision and got roughly $0.15 on Jev, against $1.48 on GPT-5 mini. The comparison that matters is not the other model though. It is MadKudu's Growth tier at $1,999 a month for up to 2,000 leads.
2. Inbox and support triage
The decision: what is this about, how angry are they, does it need a person. Same source priced 20,000 tickets a month at roughly $0.76. Zendesk Copilot lists at $50 per agent per month. For a solo founder answering their own inbox, the useful version is smaller: score every inbound email overnight so your morning starts with a sorted queue rather than a scroll.
3. Ad comment and community moderation
The decision: is this spam, a complaint, a buying signal, or a question worth answering. This is the one where volume makes the case on its own. Running a judgment on every comment across every ad set was never worth a frontier model call. At a few hundredths of a cent, you can score all of them and surface only the buying signals.
4. Creative pre-screening
The decision: does this hook match our brand, does it make a claim we cannot support, is it close to something we already ran. Developers reported using Jev to scan the Meta ad library, compare how long different ad formats stayed live, and score creative scripts before shooting, roughly 30 times faster than reviewing by hand and for under $3. Read that as a directional report from early users rather than a measured benchmark, but the shape is right: you can score fifty hooks before you commit to filming one.
5. Competitor and feed monitoring
The decision: is this post relevant to me, or is it bait. One developer filtered three days of posts with eight questions in about two seconds for $0.007. A daily competitor digest that judges relevance instead of keyword-matching is now a job you run on everything rather than a sample.
6. Fact-checking what your AI wrote
This is the one that saves the most, and it works as a cascade. Let a cheap model answer, then ask a decision model whether the answer is supported by the source it cites, and escalate to your expensive model only when the answer comes back no. On OpenRouter's published 50-question example, that pattern produced zero wrong answers, matching a frontier model on every request, at about 7% of the cost.
What the honest version looks like
The eye-catching multipliers are TypeSafe's own, and their test scores agreement with GPT-6 Astra and Claude Fable 5.1 rather than verified ground truth. On that same test Jev reached 67.8% agreement overall against 74.1% for GPT-5.6 Sol, and fell to 61.8% against 79.1% on invoice processing. Independent results are kinder and show the same trade: Dan Shipper at Every found Jev caught six of seven planted writing defects where Fable 5.1 caught all seven, at a median 0.35 seconds per passage against 8.83 and roughly 580 times lower cost.
Four things worth knowing before you move anything:
It follows lazy instructions with total confidence. "Eat" is a label. "Eat only when hunger is above 50" is a policy. Write conditions, not category names, or you will get confident nonsense at scale.
English is the primary training language, text only, no images.
It is early access with rate limits that move and pricing TypeSafe itself has not promised to hold. Build so the model is swappable.
The name is a warning. Jev is short for Jevons, the economist who noticed that cheaper coal led to more coal burned, not less. Cost per decision falls, and total spend can still rise once you start checking everything. Measure the bill, not the unit price.
Try it this week
Access opened fully today with a $5 credit, which is around 120 million input tokens at list price. It is also on Vercel's AI Gateway as typesafe-ai/jev and on Cloudflare as typesafe/jev.
Pick the single decision you make most often. Write its criteria as conditions rather than labels. Run 100 real examples through both Jev and whatever you use today, and compare them only on the cases where you would let the system act without you. That last filter is the whole test. A model that is slightly less accurate but honest about when it is unsure can automate more work than a confident one that is slightly better.
That's a wrap for this week in AI.
See you next week,
Rananjay
P.S. If you found this valuable, share it with a fellow marketer who's trying to make sense of AI. And hit reply to let me know—what's the one AI workflow you want to automate this quarter?