This Week

The Current

Issue 07 21 September 2026 9 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 Facebook put a price on posting a link, and new research found the chatbot your customers shop with talks them out of buying about as often as it talks them into it. No sign-up yet, just this week's issue.

Correction

In issues 02 and 04 we wrote that ads inside ChatGPT hadn't reached Australia, and that left a wrong impression. OpenAI extended its ads pilot to Australia, New Zealand and Canada in late March, and by May, Australian businesses could buy through agency partners or a beta self-serve Ads Manager. What we were accurately describing at the time was that the big European expansions didn't include us, but we wrote it in a way that read like the channel was shut to Australian businesses, and it wasn't. There's a fuller piece on it below.

Worth Knowing

Two basics worth fixing before you pay a platform anything

The post that tells people exactly what to do beats the clever one that doesn't

A lot of small business posts are genuinely good and then just stop, so someone watches the whole thing, enjoys it, and has no idea what you wanted from them, and I think that's where most of the wasted effort on social actually goes. People are scrolling quickly and doing three other things, so they're not going to work out the next step for themselves, which means you have to say it: book here, come in Thursday, comment the word and I'll send it, ask for it at the counter. One instruction, not three, because two asks in the same post usually means neither gets done. It feels blunt when you write it and it reads perfectly normal to the person seeing it once, and it's the single cheapest change you can make this week, given it costs you a sentence.

A short vertical video will reach people a photo never will

Every platform that matters right now is built to push short vertical video at people who don't follow you, and photos mostly get shown to the people who already do, so if you're only posting stills you're essentially talking to your existing customers and wondering why nothing new is coming in. That's not a quality judgement on your photos, it's just where the recommendation engines are pointing, and I think it's the most important structural thing to understand about organic reach at the moment. So I'd take whatever you were going to post as a picture and film fifteen seconds of it instead, on your phone, vertical, no script, just the thing happening and you saying what it is. It doesn't need to be good in a production sense, it needs to exist, because the format is doing a lot of the work here.

Facebook has put a price on posting a link

On Monday 15 September Meta launched Meta One globally, which is a paid subscription sitting across Instagram, Facebook, WhatsApp and Meta AI, with creator and business bundles starting at about fifteen US dollars a month. Australian pricing went live the same week, with the entry business tier at $22.99 a month and the top one at $769, and Meta was clear in its announcement that the core experience across its apps stays free.

Two days later the other half of it showed up. Page managers started reporting notices telling them their accounts are now limited to two posts containing an external link per month unless they subscribe, with the higher Meta One tiers unlocking more. So the free part of Facebook stays free in the sense that you can still post, and the specific thing a small business most wants to do with a post, which is send somebody to your booking page or your online shop, is the thing now sitting behind the subscription.

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

I want to be careful about what's actually confirmed here, because there's a real difference between a policy and a test. Meta has announced Meta One and its pricing itself. It has not published a two-link rule, and what we have on that side is page managers posting their notices and the trade press reporting them, on top of a smaller test that first appeared back in December 2025 and that Meta described at the time as a limited experiment. Which means nobody can tell you yet how widely this is switched on in Australia, or whether it becomes permanent, and anyone who says otherwise is guessing.

The reach on link posts went years ago. This is the invoice turning up afterwards.

What makes it sting less than it sounds is that link posts had already stopped working. Meta publishes a Widely Viewed Content Report, and its own figure for the first quarter of this year is that 98.7 percent of post views in the US contained no link to anywhere outside Facebook, so 1.3 percent did, down from 9.8 percent in 2022 when Meta first started publishing the number. Facebook has been quietly suppressing posts that send people away for years. This is the same policy with a price on it.

2022
9.8%
Q1 2026
1.3%

Share of Facebook post views in the US that included a link to somewhere outside Facebook. Meta's own Widely Viewed Content Report puts the first quarter of 2026 at 1.3 percent, meaning 98.7 percent of views contained no outside link, against 9.8 percent in 2022 when the figure was first published. Bars are drawn to scale, which is the point.

So I wouldn't rebuild anything over this, and I wouldn't subscribe to fix it either, because paying $22.99 a month to post six more links into a feed that barely shows links is solving the wrong problem. The useful read is narrower than the headlines: a link in a Facebook post was already close to worthless, and now there's a number attached to finding that out.

Do this week

Check your Page for one of these notices so you know whether you're in it, then stop putting the link in the post. Put it in the first comment or leave it in your bio, and write the post so it works on its own without anyone clicking. The bigger job, and the one worth an hour this week, is getting one line to your customers that Meta can't price: an email address, a phone number, a list of regulars you can actually message.

The chatbot your customers shop with is talking them out of things

Semrush published a survey on 18 September, run in July across 2,338 US adults with a margin of error of about two points, and the number that stopped me was this one: among people who use chatbots while they're shopping online, nearly 81 percent have decided against a purchase because of something the AI told them. Across all AI users, 58 percent have bought something a chatbot recommended and the same 58 percent have been talked out of something, which is a fairly even split for a tool most people still describe as a search box.

Swapped search
65%
Bought
58%
Didn't buy
58%
Dislike ads
42%

Share of AI users who have replaced some product-related Google searches with a chatbot, who have bought something a chatbot recommended, who have decided against a purchase based on what a chatbot told them, and who say they dislike ads inside chatbots. Semrush, survey of 2,338 US adults conducted July 2026 and published 18 September 2026, margin of error about two percentage points. Among the narrower group who use chatbots while shopping online, the figure for deciding against a purchase is nearly 81 percent.

Sixty-five percent said they've replaced at least some product searches on Google with a chatbot, which is the part that changes the job rather than just describing it. On ads inside the chatbots, 42 percent said they dislike them against 20 percent who like them, and of the people who dislike them, 66 percent said ads make them question whether the answers themselves are straight.

It isn't deciding whether to recommend you. It's repeating whatever it can find about you.

The honest caveats: this is a US sample, it's what people say they did rather than a record of what they did, and July to September is long enough for a fast-moving habit to have shifted again. But the direction matches everything else we've seen this year, and the mechanism underneath it isn't really in doubt, because a chatbot answering "is this place any good" has to get its answer from somewhere, and where it gets it is your reviews, your Google Business Profile, your website and whatever else about you is sitting on the public internet.

Which means the talking-out-of-it number isn't really a story about AI at all. It's the same old problem with a much faster delivery mechanism: three bad reviews that used to be found by someone who went looking now get summarised, unprompted, for someone who only asked a casual question. The flip side is that this is one of the few places a small business is genuinely well placed, since a real business with real customers saying real things about it is exactly what these systems have to work with.

Do this week

Open whichever chatbot you use and ask it about your own business by name, then ask it the question a customer would actually ask, something like the best place for what you sell in your suburb. Read what comes back properly, including whether you're mentioned at all. Then fix the source rather than the answer, so the wrong opening hours on your Google profile, the old menu on your website, the review you never replied to.

You can buy ads inside ChatGPT from here, and you could months ago

We've written about ads in ChatGPT twice, in issue 02 and again in issue 04, and both times the framing was that this was a channel happening somewhere else. There's a correction at the top of this issue, and this is the longer version, because the useful thing isn't that we got it wrong, it's what's actually true now.

OpenAI extended its ads pilot to Australia, New Zealand and Canada in late March, and by late May the Australian trade press was reporting that local businesses could buy either through agency partners or through a beta self-serve Ads Manager, with cost-per-click bidding added alongside the original cost-per-impression buying and measurement tools including a conversions API and a pixel. In August, an analysis reported by Mi3 found ad placements turning up in about half of the Australian ChatGPT replies it monitored, with telcos, energy and travel among the categories moving first. On 16 September the United Arab Emirates, Saudi Arabia, Israel and Turkey were added, taking the advertiser list to 56 countries, and OpenAI has said the whole thing passed a billion US dollars in annualised revenue run rate in under two hundred days.

A channel being open to you and a channel being worth your money are two different questions, and only one of them got settled this week.

So should you actually put money into it. My honest answer is not yet, and not because it's unavailable, which is the thing we had wrong, but because it's still a beta with no published small business benchmarks, the advertisers piling in are the ones who can afford to learn expensively, and on the research above, 42 percent of people say they dislike ads inside chatbots in the first place. OpenAI's own stated position is that ads don't influence the answers and that advertisers don't get access to conversations or personal data, which matters, and which is also exactly the kind of thing worth watching rather than taking as settled.

What I'd change is the assumption. If you've been mentally filing AI ads under "not here yet," that's out of date, and the practical version of that is that the traffic you might lose to chatbots answering questions for people is at least a place you could eventually buy your way back into.

Do this week

Go and create an account in OpenAI's ads manager and have a look at what it'll let you do, the same way you'd open a new ad platform for the first time. Don't move any budget into it, and don't pull any out of what's working. Fifteen minutes of looking now is worth more than reading another article about it in six months.

Google will show you what its AI was told about your business

Around 14 September Google started rolling out a new tab inside Google Business Profile called Collected Info, and it does something I don't think many business owners realise was happening at all. Google's automated assistant has been ringing and messaging businesses to confirm details, and it has been keeping what it was told, and using it to decide which searches you turn up in. This tab is the first time you get to read that back.

According to Google's own help documentation, the tab holds information collected in phone or chat conversations with your business, it's used to match you with customers looking for services like yours, and each entry shows where it came from and the date it was collected, displayed in whatever language it was given in. If something is wrong you can delete it, which removes it from Google Search and Maps, though deleting it doesn't change the other details already sitting on your profile. Google says it's rolling out to selected regions, languages and business categories, so it may well not be there when you look.

The reason I'd go and look anyway is that this is a record of what a staff member said to an automated caller on some Tuesday afternoon months ago, and then forgot about. Whether you do a particular service, whether you deliver, whether you're open on a public holiday, whether you take walk-ins. Nobody treats that call as a marketing decision at the time, and it's been quietly shaping which searches put you in front of somebody ever since.

Do this week

Open your Business Profile from Google Search or Maps while signed in as the owner, look for a Collected Info tab, and read every line in it. Delete anything that's wrong or out of date. If the tab isn't there yet, spend the five minutes checking your hours, services and description instead, because that's the same job from the other end.

If you want a say on the feed law, submissions close Tuesday at midday

Last week's lead was the exposure draft of the Digital Duty of Care bill, released on 8 September, and the part of it called My Feed, My Way, which would make platforms ask every user whether they want the algorithmic feed or a feed of only the accounts they follow. The short practical note this week is that the department has that draft open for feedback until midday on Tuesday 22 September, with submissions capped at five pages and sent by email to the digital duty of care address on the department's have your say page.

I'm not going to pretend a one page note from a Perth business changes a bill on its own, because it doesn't. But almost everything that lands in a consultation like this comes from platforms, industry bodies and law firms, and the submissions that get quoted in the end tend to be the ones from people the thing actually happens to, so there's a genuine gap there and it costs an hour to fill a bit of it.

Do this week

If this matters to you, write a page before Tuesday lunchtime saying plainly what your business would lose if most of your new customers stopped being recommended to you, with real numbers if you have them. If you'd rather not, just note that a bill is expected in Parliament later this year and nothing changes for you before then.

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

Nothing new landed at the top this week, so it's still Claude Fable 5.1 or GPT-6 Astra, and we're still not quoting an index score, for the same reason as last week: the public trackers we can see don't agree with each other right now. Claude is what we use for planning and drafting and what part of this newsletter is written with. If you already pay for one of them, that's your answer.

Best for images

Still GPT Image 2.5 and its Flare and Sunburst variants, unchanged from last week, and we'd rather say that than find a new winner every seven days. The advice hasn't changed either: the useful half of it for a small business is editing a real photo of your real thing, not generating a convincing photo of something that doesn't exist.

Best for video

Third week running with nothing at the top worth switching for, so pick on price, on how long a clip you actually need, and on whether it can do your specific shot. The one dated thing here: the Sora API shuts down this Thursday, 24 September. If you pay for a video tool that generates its clips through Sora, and plenty of smaller ones do, ask them today rather than on a day you need a video.

Worth flagging

Ads inside ChatGPT reached 56 countries on 16 September when the UAE, Saudi Arabia, Israel and Turkey were added, and OpenAI says the business passed a billion US dollars in annualised run rate in under two hundred days. Australia has been able to buy since about March, which we got wrong twice and have corrected at the top of this issue. Worth looking at, not worth moving budget into yet.

We got one wrong, so here's how we handle that

There's a correction at the top of this issue, and since nobody reads this newsletter because we're infallible, I'd rather say something about it than let it slide past. We told you twice that ads in ChatGPT weren't available in Australia, and what we were actually looking at was American and European coverage of expansions that didn't include us, which is a different thing to the channel being closed here. The lesson we've taken is that a newsletter written from mostly American trade press will quietly default to an American view of what exists, and for a Perth audience that's not good enough, so we're going to check the Australian trade press directly on anything we say isn't available here. Past issues don't get edited, because they're a record of what actually went out, so corrections run here in the next one instead. 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, 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.