This Week

The Current

Issue 04 31 August 2026 6 minute read Past editions
A laptop screen showing marketing analytics
Photo by ThisIsEngineering via Pexels.

If social media for your business feels like a moving target, this is for you. Plain, honest notes each week on what's actually working, from two people who run it for a living. This week is mostly about the platforms deciding things on your behalf. No sign-up yet, just this week's issue.

Worth Knowing

Two things worth getting right before anything else

Most people are watching with the sound off, so the captions are doing the work

People scroll in bed next to someone asleep, in a queue, at work, on the train, and a lot of them never turn the sound on at all. So if the thing you need someone to understand only exists in the audio, most of your audience never got it. The fix isn't complicated, you just say the important part on the screen as well as out loud, in the first second or two, in text big enough to read on a phone held at arm's length, and you let the words carry the meaning while the sound carries the mood. I think this is the single cheapest improvement most small businesses could make to what they're already posting, because it costs you nothing and it doesn't require you to be any better on camera.

The fastest way to reach new locals is usually another local business

If you're small and local, the people you want are already following somebody near you who isn't your competitor, so the cafe, the physio, the gym, the florist, the bloke who does the mobile detailing. Doing something together, a joint post, a giveaway, filming in each other's space, one of you genuinely recommending the other, puts you in front of an audience that's already local and already trusts the person introducing you, and that's a much warmer introduction than an ad ever gives you. It costs nothing but a conversation, and I think it's underused mostly because it feels like asking a favour, when in practice you're offering one, since they need content as much as you do.

Meta is rebuilding teen accounts, and it starts in the US

On 27 August Meta agreed to pay up to $18 billion to settle claims brought by attorneys general from nearly every American state, over the argument that Facebook and Instagram were built in ways that harmed young users. The money is the headline everywhere, but the part that actually changes the product is the list of things Meta agreed to switch on by default for accounts belonging to under-18s.

Daily capNone2 hours
OvernightOpenBlocked
School hoursNotifiedMuted

Default settings for under-18 accounts on Facebook and Instagram, before and under the settlement announced 27 August 2026. The daily cap covers the two apps combined, and the overnight block runs midnight to 6am with direct messages excepted.

The two hour limit and the overnight block can only be loosened with a parent's permission rather than by the teenager, teens get nudged every fifteen minutes of continuous use and again after an hour, and like counts are hidden. Bloomberg reported the same day that all of this applies to users in the United States only for now, which is the part most of the coverage skipped past.

The rules changed in America, and nobody has said when or whether they change here.

We're not far from our own version of it either. On 19 August the eSafety Commissioner said platforms are likely to face enforcement action over Australia's under-16 ban, with Facebook, Instagram, Snapchat, TikTok and YouTube all named as compliance concerns, and eSafety's own first evaluation put the fall in under-16s holding accounts at about 20 percent, which reads to me like the ban is working slowly rather than not at all.

If your customers are adults, and for most of the businesses I talk to they are, none of this changes what you do on Monday. What I'd watch is the second order effect, because you can't enforce a rule about teenagers without checking everybody's age, so age assurance is quietly becoming a normal part of signing up to and staying on these platforms, and every extra step between a person and your content is friction that wasn't there a year ago.

Do this week

So there's nothing to action this week, just something to file. If any part of your audience is under 18, or you sell to parents on their kids' behalf, this is the year to build a way of reaching people that you actually own, an email list or a phone list, because the platforms are increasingly being told who they're allowed to show things to.

Google will now spend up to your target instead of under it

This one took effect on 17 August and I don't think many small advertisers noticed, partly because Google announced it back on 15 June and partly because it isn't the sort of change that turns up in your inbox. If a Google Ads campaign is limited by its budget and it runs on a target based bidding strategy, meaning target CPA or target ROAS, it will now perform more consistently toward the target you set instead of quietly beating it.

Cost / lead$5Toward $10

Google's own worked example: a budget limited campaign with a $10 target cost per acquisition that had been delivering at $5 will now deliver closer to $10. Announced 15 June 2026, in effect from 17 August 2026.

The way I'd explain it is that your target used to behave like a ceiling you rarely touched, and now it behaves much more like an instruction. Google would say the campaign is simply doing what you told it to do, and that's fair enough, except that a lot of people set that number as the most they could stomach on a bad month rather than the number they actually wanted to pay, and there's a real difference between those two things once the platform starts taking you at your word.

A target used to be a ceiling you rarely touched. It behaves like an instruction now.

It only affects campaigns that are limited by budget, and if you're running a small daily spend on Search, Shopping, Performance Max or Demand Gen then that's almost certainly you. Campaigns that aren't budget constrained carry on exactly as they were. What I can't tell you is how big the swing will be in your account, because it depends entirely on how far under your target you'd been sitting, and there's no independent data on it yet.

Do this week

Open the account, look at the last month, and put your target CPA next to what you were actually paying. If there's a wide gap, bring the target down near what you were really getting instead of the number you set once and forgot about. Then write today's date on it, because if your cost per lead climbs over the next few weeks you'll want to know whether it was this or you.

You can swap the audio on a post you have already published

Quietly useful, and I've barely seen anyone mention it: Instagram added a replace audio option to feed posts and carousels in late July, so you can change the track on something that's already live rather than deleting it and putting it back up. You tap edit on the post, choose a different track, and it keeps everything it had already earned, the likes, the comments, the reach.

Two limits worth knowing before you go looking. You can only pick from Instagram's own audio library, so you can't attach a file of your own to a post that's already live, and Instagram framed this around posts and carousels rather than Reels, so I'd treat Reels as unchanged until you actually see the option sitting there.

A person checking a social media grid on their phone next to a laptop
Photo by Plann via Pexels.

The reason this matters more than it sounds is that audio is one of the only reach levers that still works after you've hit publish. A trending sound is a distribution signal rather than a message, so it can push an old post back in front of people who never saw it, while the actual meaning of the post still has to survive with the sound off, which is how a good chunk of your audience will watch it either way.

What nobody has published is any data showing that swapping the audio genuinely revives reach on an old post, so I'd treat this as a cheap experiment rather than a strategy, and I'd be suspicious of anyone telling you otherwise this early.

Do this week

Pick two or three older posts that did well, put a current track on them, write down the date you did it, and look again in a fortnight. It costs you about ten minutes, and if nothing happens you've lost about ten minutes.

ChatGPT ads just landed in 31 more countries, and none of them are here

OpenAI turned ChatGPT ads on across 31 European markets in the week of 24 August, the 27 EU countries plus Iceland, Liechtenstein, Norway and Switzerland, and it's the biggest expansion its advertising business has had so far. Australia isn't in it, so there is nothing here for you to buy this week, and I want to be upfront that this is a story about direction rather than a story you can act on.

The details worth noticing are the ones about who gets in. European advertisers are starting through OpenAI's own ads team and its agency and technology partners rather than through a self-serve tool, with self-serve access said to follow, so for now it's built for people with a media buyer rather than for a business with a card and a Tuesday afternoon. The ads only show to people on the free and Go plans, not to anyone paying for Plus, Pro, Business, Enterprise or Edu, and OpenAI says they aren't shown to accounts it identifies as belonging to under-18s.

The question stops being where you rank, and becomes whether you get mentioned at all.

The reason I keep flagging this even though you can't do anything with it is that it changes the shape of the thing rather than the price of it. If a decent chunk of people start asking an assistant for a recommendation instead of searching and scrolling, then what matters isn't where you sit on a page of results, it's whether you get named inside the answer, and ads are just the first part of that which somebody worked out how to charge for.

I'd also say plainly that nobody outside OpenAI has published anything credible on whether these ads work, so anyone selling you a ChatGPT ads strategy right now is selling you a guess with a nice deck around it.

Do this week

Nothing to do, other than the thing that's been quietly true for a while: the businesses that get named inside AI answers tend to be the ones with real reviews, a website that says plainly what they do and where they do it, and actual people talking about them somewhere public. That's the same work that helps you anyway, which is why I'm comfortable telling you to do it before anyone can prove it pays.

What's actually good right now

Disclosed upfront: this newsletter is partly written with Claude, so take our AI opinions here as disclosed rather than neutral.

Best all-round assistant

No change worth reporting this week. Claude Opus 5 still sits on top of Artificial Analysis's Intelligence Index at about 63, and it's what we use for planning and drafting. These rankings move by a point or two constantly, so I wouldn't switch tools over a decimal place, I'd switch over whether the thing is actually helping you.

Best for images

GPT Image 2 still leads Artificial Analysis's text to image arena at roughly 1370 Elo, about 97 clear of second, and it stays the one to reach for when the words on the image have to be readable. Microsoft announced MAI-Image-2.6 on 10 August claiming second place and much better text rendering, so it's worth a look. For changing an existing photo rather than making one, Nano Banana Pro still leads the editing arena.

Best for video

Honestly, the leaderboards are contradicting each other right now, with different arenas putting different models first depending on whether the clip has sound and whether you started from text or a photo. When the rankings disagree that much, the ranking isn't the useful input, so choose on price, clip length and whether it can do the specific shot you need.

Worth flagging

The detection company Pangram published a study in July, across just over a million posts, estimating that 41 percent of long form LinkedIn posts now read as entirely AI written, against about 13 percent on Reddit and 10 percent on Substack. Their scores are estimates from writing style rather than proof of how anything was made, so hold the numbers loosely, but the direction is the point: the feeds are filling up with competent, weightless writing, and sounding like an actual person with an opinion is becoming the thing that stands out rather than the thing you apologise for.

Set the target from your margin, not from what the platform will tolerate

The Google change further up the page is small on paper, and I think it's a good example of something worth being deliberate about, so here's how we'd rather do it. A target cost per lead should come out of what a customer is genuinely worth to you, so you work back from your margin and the share of leads you actually close, and you land on a number you can defend to yourself. What tends to happen instead is that the number gets set to the most the business could bear on a bad month, and then everybody forgets it's sitting there, which was fine while the platform quietly came in under it and is less fine now that it doesn't. On our own numbers we're still light, since the case studies are being built out of two years of Dr Crack Wax content and aren't finished yet, so when we tell you something works we'll keep being clear about which part is our data and which part is somebody else's.