ChatGPT Is Two Weeks Old. Some Guesses for 2023

Posted by Sean on December 14, 2022

OpenAI released ChatGPT on 30 November. By the following Monday its chief executive was saying it had passed a million users. I’ve lost count of the screenshots in my feeds: poems, cover letters, recipes, code, people trying to get it to say something it shouldn’t. Stack Overflow has already put a temporary ban on answers generated by it, because they were arriving in volume and too many were confidently wrong.

I usually write some predictions in December. This year it would be odd to write about anything else first, so here’s where my head is. Treat all of it as guesswork two weeks into something new.

What it’s actually good at

From what I have seen of it so far, it’s better than anything I’ve used at producing fluent text on request. Ask it to explain GA4 data retention to a finance director and you get a tidy, plausible paragraph. Ask it for twenty product name ideas and you get twenty, instantly.

It’s also wrong a lot, and wrong in the same confident tone it uses when it’s right. Ask it for sources and it will sometimes invent them. That combination is the thing I’d want every marketer to understand before they let it near anything that gets published.

Prediction 1: a flood of content, fast

The AI writing tools built on GPT-3 were already producing blog posts at volume. ChatGPT makes it free and easy for anyone. I expect a lot of sites to start publishing far more articles in the first half of 2023, much of it lightly edited machine output.

Google’s position, as far as I can tell, is that automatically generated content meant to manipulate rankings is against its spam policies, and August’s helpful content update asked directly whether sites were using extensive automation to produce content on many topics. So I’d expect a year of cat and mouse: lots of output, followed by some sites being hit hard. I wouldn’t want to be on the wrong side of that with an ecommerce site that depends on organic traffic.

Prediction 2: people start asking it instead of searching

I’m least sure about this one. For some questions, “how do I convert a recipe for a fan oven”, “what’s the difference between worsted and woollen yarn”, a chat answer is quicker than ten blue links. If enough people get used to that, some informational search volume will move.

I don’t think it threatens Google in 2023. ChatGPT doesn’t know about anything recent, can’t tell you who has something in stock, and doesn’t show sources. But I’d be surprised if Google isn’t working hard on its own version, and I’d watch for chat-style answers turning up in search itself. Informational content that exists only to catch a question and send the visitor on is the most exposed.

Prediction 3: useful in the boring places first

The marketing value I can see right now is in tasks nobody enjoys. First drafts of product descriptions from a spec sheet, for someone to then correct. Rewording the same delivery message for twelve contexts. Summarising a long customer feedback export. Writing a regex for a GA filter. It saves time, provided someone checks the output.

Prediction 4: a lot of policy written in a hurry

Businesses will need to decide what staff can paste into it. Customer data, sales figures and unreleased product plans going into a free tool on someone else’s servers should worry anyone responsible for data protection. I expect a lot of companies to write a quick policy in January, and some to find out in February that people were already doing it.

The usual ones

So I don’t forget them:

  • GA4. Universal Analytics stops processing on 1 July. If you’re not collecting data in GA4 by now, you won’t have a full year-on-year comparison when it goes. Start this week.
  • Cookies. Chrome’s third-party cookie phase-out is now pencilled in for the second half of 2024. Use the extra time for first-party data rather than relaxing.
  • Costs. With inflation where it is, I expect a lot of marketing budgets to be cut or frozen in the first quarter, and more scrutiny on paid media returns.

I’ll look back at these next December and see how badly I did.

← Previous Draft Draft Index Next Draft →