A huge week for OpenAI on the product side. Images 2.0 has landed and ChatGPT 5.5 followed close behind.

The image generation upgrade is the one I've explored most. I run image generation at scale through Kie.ai with Claude orchestrating the workflow, and the lift over the past week is significant. Better quality, faster output, and the cost is still reasonable for production use.

This is the OpenAI I'm interested in. The product team shipping things people actually use, while the commercial side figures out ads and the leadership figures out what business they want to build.

Images 2.0 isn't a side quest, to use OpenAI's own internal phrase. It's a real reason to keep paying for the platform if you're building anything that needs visual output at volume.

If you've not tested it yet, run a comparison. Same prompt, same brief, side by side with whatever you were using before. The gap in quality is huge.

A lot of people still dunk on vibe coding. I get it. The output of someone prompting their way through a build with no engineering instinct can be rough (and is dangerous if you’re selling this into clients).

The barrier between "I have an idea" and "I have a working version of that idea" has thinned out. People who would never have shipped before are shipping. Founders without engineering teams are wireframing, building landing pages, and prototyping flows that would have needed a full sprint six months ago.

For Obsero, I've got Claude plugged into Pipedrive, PostHog, and Claude Design (rolled out last week). It eats into the usage limits, but the output justifies it. I can draft a wireframe, pull live trial data into the conversation, and have the recommendation back in front of the team before lunch.

This week Meta announced it's laying off 10% of its workforce as it leans further into AI. That's a tough headline if you're inside one of those teams. I don't always think redundancies are a bad thing, but it is a concerning time unless you embrace the change.

The same week, ChatGPT 5.5 and Images 2.0 ship. Claude Design exists. MCPs let any non-engineer plug their tools into a model that can actually do the work. If you're a marketer, a founder, or a consultant, this is the most leveraged moment in the history of the discipline. The cost of starting something has never been lower. The cost of not learning these tools has never been higher.

There's also been a lot of pushback against using AI for content and creative work. Fair, when the output is slop. Less fair when people assume that's all it can be. If you ask ChatGPT to write a newsletter without plugging in your systems, your data, your voice, or any review process, you'll get exactly what you asked for. That's a workflow problem, not an AI problem.

I've seen LinkedIn posts and YouTube tutorials pitching "automate your newsletter forever" workflows and I genuinely don't understand why you would do this. If your newsletter exists to capture ad revenue without you ever thinking about it, what are your subscribers actually getting?

The reason I write Franc Talking is to force myself to reflect on the week. The constant change in this category means you have to keep up, and writing about it is how I keep up. Mass producing content for ad revenue without a point of view is the path to becoming part of the slop you're complaining about.

This has bugged me from the start.

I've been in SEO for over 20 years. Led teams at Cazoo, P&G, and IG Group. I've watched the discipline change more times than I can count, and I don't understand the negativity towards GEO right now.

The energy spent arguing about whether Generative Engine Optimisation is a real thing is energy not spent helping clients adapt. Every hour spent dunking on (genuine) GEO consultants on LinkedIn is an hour not spent figuring out why your client's brand is missing from ChatGPT answers in their core category.

If you've built a strong relationship with a client over years, the chance they'll drop you for a GEO consultant who walked in last week is small. What's far more likely is they'll ask you a question. Do I need to care about ChatGPT? Should I be in AI Overviews? How do I measure this? You either have a good answer, or someone else does.

The best response is to broaden your expertise. Talk about how AI search works. Explain why training data matters. Explain that getting cited once isn't the goal. Explain what deterministic vs probabilistic means for answers. Being cited consistently, across the prompts your customers actually use, in a way that reflects your brand the way you want it reflected, is the goal. That takes time, content investment, and consistent signals across the web. It's a brand building exercise, not a campaign.

That's the bit the 'snake oil salesmen' framing misses. If you're serious about AI search, you can't shortcut the foundations. Training data is the foundation of everything an LLM tells a user about your brand. You don't influence training data overnight. You build a brand that's worth being cited, you create content that earns its place, and you do that consistently for long enough that the models reflect you the way you want to be reflected. That's brand building. Not a campaign you run for a quarter and walk away from.

Spending less time arguing about names and more time helping clients understand what's actually changing would do the industry a lot of good.

I'll be at Brighton SEO on Friday. Mike is there both Thursday and Friday. If you're attending and want to talk AI search, GEO, or what we're building at Obsero, drop me a message.

My old mentor Jon Earnshaw, Chief Product Evangelist at PI Datametrics, is on stage. Jon was instrumental in shaping how I think about search early in my career and it's always good to hear him talk.

I expect the agenda to lean heavily into AI search. Query fan out, LLM tracking, prompt visibility, and the inevitable round of debates about what to call all of it. I'm interested in hearing how different categories are approaching this. B2B and ecom are at very different stages of maturity in AI search, and the gap between them is getting wider, not smaller.

If you're going, come and say hello.

Last week I covered OpenAI's ads pixel hitting the wild and the move from $200k upfront commitments to $30-50k monthly deals. This week another piece dropped: which I first saw here.

ChatGPT ads now support CPC, with max bids of $3-5. The platform was previously running on impressions, which is fine for awareness but useless if you're trying to prove ROI. CPC is the model every performance marketer expects, and rolling it out tells you OpenAI is serious about competing for direct response budgets, not just brand spend.

This matters because the targets they've set are aggressive. $2.5bn in ad revenue by the end of 2026 and $100bn by 2030. You don't hit numbers like that on impressions alone. You hit them by giving advertisers tools they recognise from Google Ads and Meta, with attribution they can defend to a CFO.

Tracking and conversion measurement are still patchy inside ChatGPT, but the speed of iteration is striking. This is a company that only became a household name in Q4 2022. They're now building an ads platform at the pace Facebook took five years to build. I'd expect every piece of the standard performance marketing stack to land inside ChatGPT over the next 12 months.

If you're a brand that hasn't yet thought about ChatGPT as an advertising channel, this is the moment to start. Not necessarily to spend, but to understand the format, the bidding mechanics, and where it sits in your funnel.

In 2026, it’s easier than ever to start a newsletter about a topic you love, grow your audience, and increase revenue.

Whether you’re in B2B SaaS, FinTech or want a side hustle, consider starting a newsletter today and beehiiv is the perfect partner for this.

Use my link here and start your 14-day trial today. If you sign up, you get 20% off your subscription fee for 3 months!