This Week

The Current

Issue 09 5 October 2026 8 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 Meta gave its AI agent a small business version and didn't bring it here, new research found that over Christmas it's how often you email people that costs you their trust rather than what you send, and TikTok told a room in Sydney that Australian brands have doubled what they spend advertising overseas.

Worth Knowing

Two basics sitting underneath this week

Tell the AI what you want the way you'd tell a new staff member

The single biggest difference between people who get something useful out of these tools and people who decide they're overrated is how much they put into the ask, so if you type "write me an Instagram caption" you will get the beige thing everybody else got, and if you type out who you are, who's reading it, what you're selling, what you sound like and what you want the person to do next, you'll get something you only have to tidy. I think the useful mental model is that you're briefing somebody on their first day rather than searching, because a new staff member who knew nothing about the business would need all of that from you too, and nobody would be surprised when they got it wrong without it.

Three or four things you post about beats a fresh idea every time

Most businesses that stop posting don't stop because the content failed, they stop because every single post started from a blank page and that's exhausting in a way that nothing else about running a business is, so the fix is to decide on three or four subjects you'll come back to forever and then just rotate them. For a cafe that might be the people who work there, how a thing on the menu gets made, regulars and what they order, and whatever's changed this week, and the point isn't that those are clever, it's that you never again have to answer the question of what to post, only which of the four it is today.

Meta built a small business version of its AI agent, and you can't have it

On Tuesday 29 September Meta announced Muse for Small Business, which is a version of its AI agent Muse pointed at people running a business rather than people planning a holiday. Muse itself is new, it arrived in the United States on 8 September and in Canada on 18 September, and the thing that makes it different from a chatbot is that it does things instead of answering questions: it works across apps, browses, fills things in, and asks you to approve steps as it goes. What Meta added on Tuesday is the business plumbing, so you can connect your Instagram professional account analytics, your Facebook Page and your Meta ad accounts to it, plus a long list of outside tools including Canva, Shopify, Intuit QuickBooks, Klaviyo and Slack, and then ask it to work across the lot. Meta's own line was "Give Muse a goal, like running your business or finding new customers, and it gets it done," and the jobs it listed were analysing how the business is doing, drafting campaigns and customer replies, flagging problems, and making recommendations off whatever you've plugged in.

Now the part that matters most and is the reason this is written the way it is: you cannot use it. Muse is live in the United States and Canada and nowhere else, and if you sign up from here you get told it isn't available in your country or region yet, with no date attached to that. So this is not a thing to go and try on Monday, it's a thing to understand is coming, and I'd rather tell you that plainly than write five hundred words that quietly imply you can.

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

The version of this that eventually reaches Australia will not ask whether you were ready for it.

On price, Muse is free with usage limits and there are paid plans at twenty and a hundred US dollars a month above that, and Meta didn't publish separate pricing for the business version, so the honest answer on what this costs an Australian small business is that nobody knows yet. Meta did something smaller in the same direction inside Meta AI back in August, which we covered in issue 03, and Muse is a separate app rather than that one renamed, though Meta hasn't really spelled out how the two are meant to sit next to each other.

What I think is actually worth taking from it has very little to do with the product. For as long as any of us have been doing this, your ad account and your Page and your Instagram insights have been dashboards, things you open and look at and draw a conclusion from. Meta is now treating them as inputs that something else reads and acts on, and every platform is moving the same way at once, so the accounts you've half neglected are about to be the raw material an agent works from. A Page with the wrong hours on it and an ad account full of abandoned campaigns from last year were untidy before. They're about to be instructions.

Do this week

Nothing to buy and nothing to sign up for. Spend twenty minutes making your accounts readable: check your Facebook Page and your Instagram professional account actually carry your current hours, suburb, services and a working contact method, and go through your Meta ad account and archive the half-finished campaigns you're never switching back on. That's worth doing this week whether or not an agent ever turns up, because a customer reads the same information.

Over Christmas it's how often you email, not what you send

Validity published its Holiday 2026 Consumer Survey on Wednesday 30 September, a survey of a thousand US consumers about what makes them open, ignore or unsubscribe from the emails businesses send them over the Christmas run, and the finding underneath all of it is that the damage comes from frequency rather than content. Forty four percent said they get more promotional email over the holidays, which is not surprising, but thirty six percent said that volume has actually made them miss, delete or overlook something that mattered and wasn't promotional, and about a third said the volume on its own is enough to make them unsubscribe. Not a bad subject line, not a boring offer, just too many.

Trust
47%
More email
44%
Discount
39%
Missed one
36%

Share of consumers who say already trusting a brand factors into whether they open its holiday email, who say they receive more promotional email over the holidays, who say a discount or promotion factors into opening, and who say holiday email volume has caused them to miss, delete or overlook an important non-promotional message. Validity Holiday 2026 Consumer Survey, 1,000 US consumers, published 30 September 2026.

The two bars worth putting side by side are the first and the third. Forty seven percent said that already trusting a brand is part of why they open its holiday email, ahead of the thirty nine percent who said a discount or a promotion is, which means the thing doing the work in a crowded inbox in December is who you already are to that person rather than how big the number on the offer is. Validity also found that nearly one in four people overall, and nearly half of Gen Z, say a brand's holiday email habits have made them trust it less, buy from it less or stop buying from it altogether, and that is a real cost that doesn't show up anywhere in an open rate.

A third of people will leave over how often you turn up, before they've read a word of what you sent.

The caveats, because they matter here. This is a thousand US consumers, so the sample is small and it isn't Australian, and American "holidays" means the Thanksgiving to Christmas stretch, which maps onto our Christmas trading period closely enough to be useful but isn't the same thing. It's also what people say they do rather than a record of what they did, and people reliably say they hate marketing more than their behaviour suggests. What I'd still take from it is the direction, which matches what anyone who has watched an unsubscribe list in December already suspects.

One small, practical thing in there that I hadn't seen framed this way: twenty two percent said a temporary snooze or "opt down" option would help them manage the volume without unsubscribing outright. Most small businesses don't offer that and it costs nothing to add a line to a Christmas email saying people can reply and you'll take them off the list until February, so you keep the address instead of losing it permanently over four weeks of noise.

Do this week

Before you build the Christmas calendar, write down the actual number of sends you're planning between now and Boxing Day and look at the number rather than the plan. If it's more than one a week, cut it back and put the effort into the ones that are left, and add a line offering people a break until the new year instead of an unsubscribe. The same question is worth asking of your posting schedule, though be clear that this research is about email and I'm not going to pretend it measured anything else.

Brands say creators are paying off, and that AI is what could wreck it

CreatorIQ published its seventh annual State of Creator Marketing report on Wednesday 30 September, built on a survey of two thousand brand and agency marketers worldwide run by Sapio Research across July and August. Forty eight percent of brands now report three times or better return from creator marketing, up from thirty seven percent a year ago, and budgets are up thirty three percent year on year, so the money is moving and the people spending it say it's working. The part I found more interesting is what those same people named as the biggest threat to trust in the whole channel, which was AI: AI generated content, AI influencers, and creators not disclosing either, cited by forty four percent. Forty five percent also said AI has made creator marketing more complicated to run day to day, against thirty eight percent who said it made it simpler.

I want to be upfront that this is a survey of brands with budgets and the agencies they hire, which is not you, and a two thousand person global sample of marketers describing their own results has an obvious flattering bias built into it, since nobody enjoys reporting that the thing they spent the budget on didn't work. So don't read forty eight percent at three times return as a number you should expect.

What the money is actually buying is that a real person was willing to put their name on it.

The useful read for a small business is in the gap between those two findings. The channel that's growing fastest is the one where a recognisable person vouches for something, and the thing the people funding it are most afraid of is the cheap synthetic version of exactly that. That lines up with the research we covered in issue 06, where Clutch found fifty three percent of people said they'd be less likely to buy from a brand they knew used AI in its social content, and it's the same finding arriving from the other side of the table: consumers say it costs trust, and now the marketers say it's the main risk to the channel.

For a local business the version of this you can actually run isn't creator marketing in the sense that report means. It's a regular customer with two thousand followers in your suburb, a tradesperson you already work alongside, a staff member who's good on camera, and the reason that works is the same reason the budgets are growing, which is that it's visibly a person and not a brand talking about itself. That costs a coffee and an hour rather than a line in a media plan.

Do this week

Write down five people who already post publicly about the kind of thing you sell and who actually live or work near you, then ask one of them in for something free in exchange for them filming whatever they honestly think of it. Don't script it and don't ask for approval over the edit, because the whole value is that it reads as a real opinion. And wherever you do use AI in your own content, say so, since the research on both sides of this now points the same way.

Australian brands doubled what they spend advertising overseas on TikTok

TikTok ran an event called Take Off in Sydney and put numbers out around it on Wednesday 30 September, reported here by Campaign Brief on 1 October. More than one in three major Australian and New Zealand advertisers on TikTok are now putting money into markets outside this one, and over the past year AUNZ brands have more than doubled their TikTok ad investment into international markets, up a hundred and five percent year on year, mostly aimed at the United States, the United Kingdom and Europe. TikTok also said the number of active small and medium business advertisers in Australia grew forty five percent year on year between 2024 and 2025.

If you run a cafe in Vic Park or a gym in Joondalup then selling to Britain isn't the point and I'm not going to pretend it is. There are two things in here worth a minute anyway. The first is that forty five percent figure, which is about the auction you're already bidding in rather than about exports, because more Australian small businesses advertising on TikTok means more competition for the same attention, and that's the sort of thing you feel as a slowly rising cost rather than as an announcement. It's TikTok's own number and it covers 2024 to 2025 rather than right now, so take it as direction and not precision.

The second is that if what you sell fits in a box and you can post it, trying another country on TikTok is a setting in a campaign rather than a shopfront and a visa, and the hundred and five percent is other Australian businesses working that out. That is genuinely different from how this used to work. TikTok Shop, for what it's worth, still isn't available in Australia.

Do this week

If you're local only, nothing, and you can skip this one with a clear conscience. If you already ship interstate and your product travels, set up one small campaign targeting a single overseas city at a budget you'd be relaxed about losing, run it for two weeks, and compare the cost per result against what you pay here before you form any opinion about it.

Instagram has finally said how long a Reel should be

On Monday 28 September Instagram published its own step by step guide to making Reels, covering recording, editing, sharing and collaborating, and tucked inside it is an answer to the question I get asked more than any other. Instagram says a Reel can run from three seconds to three minutes, and then gives its own ranges for what to use when: three to fifteen seconds for quick transitions, trends and one liners, fifteen to sixty seconds for tutorials, day in the life videos and most ordinary things, and sixty seconds to three minutes for longer stories, walkthroughs and detailed how to videos. For anyone starting out it says to begin at fifteen seconds, because short ones are easier to edit and more likely to get watched all the way through.

The reason this is worth a story rather than a footnote is the thing it quietly kills. There's a persistent belief that there's a secret optimal Reel length, seven seconds or twenty three seconds or whatever the current number is, and that hitting it unlocks reach. The platform's own advice is the boring version instead: match the length to what the video is actually doing, and go short while you're learning because finishing matters more than the clock does.

Be clear about what this is though, because a guide is not an algorithm disclosure. Instagram is telling you what tends to work, it isn't publishing how Reels get ranked, and nothing in it says a fourteen second Reel outperforms a sixteen second one. What it does give you is the platform's own framing, which is more than you get most weeks, and it's a reasonable thing to point at the next time somebody tells you with great confidence what the magic number is.

Do this week

Take your next three posts and decide which band each one sits in before you shoot anything: the quick visual one at fifteen seconds or under, the explaining one somewhere under a minute, the proper walkthrough longer if it genuinely needs it. Deciding first is the whole trick, because the version where you shoot three minutes of footage and then try to work out what it was for is how most of them end up too long.

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

Two of them in two days. On Monday 28 September Anthropic released Claude Sonnet 5.5, the cheaper everyday model that sits beside Opus 5.5, at the same price as Sonnet 5 and with Anthropic saying output is more than thirty percent faster and up to thirty percent cheaper per task in its own testing. On Tuesday 29 September at its DevDay in San Francisco, OpenAI released GPT-6.1 Sol, which it describes as close to the capability of its top model at a fifth of the price. We're still not quoting an intelligence index score, fourth week running, because the public trackers we can see disagree with each other and with the numbers we printed in issue 05, and we'd rather say that than pick the one that suits. For drafting and planning, either is fine.

Best for images

Still GPT Image 2.5 and its Flare and Sunburst variants, sitting top of Artificial Analysis's text to image arena, and that's a repeat of the last three weeks which we'd rather say than invent a new winner every seven days. Given what's in this issue about creators and trust, the half of image AI worth your time is still editing a real photo of your real place rather than generating a fake one of somewhere that doesn't exist.

Best for video

Nothing changed this week, which is the fourth week in a row we've written that, and we're still not naming a leader because the public arenas we can check disagree with each other about who's on top. If you want AI involved in video right now, the dependable use is still cutting and captioning something you filmed, not generating the footage.

Worth flagging

Look at the shape of the week rather than any one release. Meta shipped an AI agent aimed at small businesses that plugs into ad accounts, and OpenAI's DevDay was mostly about agents that keep running in the background, including a thing called Dots, rather than chatbots you prompt and wait on. The common thread is software that acts on your accounts instead of handing you an answer to act on yourself, and whatever you think of that, it's the direction every one of these companies is pointed at the moment.

We are not handing the account over

Half of this issue is software offering to run the thing for you, so it's fair to say where we sit on it. We use AI constantly, for drafting, for getting a long edit down to something watchable, for arguing with ourselves about what to test next, and this newsletter is partly written with Claude and says so every week in the AI Corner. What we haven't worked out how to hand over is the part where you stand around in somebody's shop for two hours and notice the one thing that's actually worth filming, because that's where nearly every post that has worked for a client has come from, and no agent has ever been in the room. I think the honest version of where this goes is that the admin gets automated and the noticing doesn't, and if that turns out to be wrong we'll say so here rather than quietly change the story. Our own numbers are still thin either way, we're a young studio and the case studies are still building out, so treat anything we claim about what works as us showing our working rather than something proven at scale.