This Week

The Current

Issue 12 26 October 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 the same thing happened three times over: Google can now go and find videos for your ads by itself, OpenAI starts rewriting your ad copy on Thursday whether you asked for that or not, and Meta's own numbers say ads got 14% dearer almost everywhere except roughly where we live.

Worth Knowing

Two basics underneath all of this

The first second is the one part no tool can fix for you

People decide whether to keep watching in about the first two seconds, which is why the opening frame and the opening line carry more weight than everything after them put together. I keep coming back to this one because it's the only part of a post that software can't repair later. An ad platform can now resize your image, rewrite your headline, pick a different video and send the click somewhere else, and none of that touches the question of whether the first second gives somebody a reason to stay. So if you only get to be deliberate about one thing in a post, be deliberate about how it starts: show the thing, say the problem, or move, and save the introduction for people who are still there at second ten.

Social gets you met, the catching has to happen somewhere you own

Social media is mostly the top of the funnel, it's how people find out you exist, and it rarely closes anything by itself, so the thing that actually turns attention into money is whatever sits on the other side of the tap. I think that matters more this year than it used to, because the platforms are getting more comfortable deciding where a click lands and what the ad said on the way there. The page, the booking form, the phone number and the follow up are the parts that are still entirely yours, and they're worth the same care you give the content. A post that reaches ten thousand people and lands them on a page with last year's prices on it has cost you more than a quiet week would have.

Google can now go and find videos for your ads, starting with your own old posts

On Thursday 23 October, Barry Schwartz at Search Engine Roundtable spotted a new on and off switch inside Performance Max campaigns called Sourced videos, sitting in the Asset optimization section and marked New. Search Engine Land and PPC Land both covered it the same day. What it does, in Google's own description of the setting, is use Google AI to identify relevant videos from your approved sources and add them to your campaign automatically, and the reported ceiling is up to ten crawled videos per asset group.

The approved sources are the part worth slowing down on, because they're not a stock library. They're your own things: your social media accounts, the landing page the ad points at, and the YouTube channel linked to the account. So the pool Google draws from is whatever you have already posted, including the stuff you posted in 2023 and forgot about.

Your ad creative used to be the list of things you uploaded, and from here it's everything video shaped that Google can reach.

On the default, I want to be careful, because the reporting doesn't fully agree and Google hasn't said anything at all. Google's help page for sourced videos describes it as enabled by default for all new campaigns and says it may have been activated automatically on some existing campaigns as part of a recent platform update. Some write-ups read that as on by default everywhere, others as a change from an older opt-in setting, and Search Engine Land's advice is the sensible one: go and look at your own account rather than assume it's off. There's no Google blog post, no changelog entry and nothing from the Ads Liaison account on this, so everything here rests on the help documentation and the people who found the toggle.

One reported detail I'd flag as reported rather than confirmed: a couple of the write-ups say videos Google sources for you get uploaded to a Google-owned YouTube channel carrying your logo and business name, and that campaigns need a daily budget above about five US dollars to be in it. Google hasn't published either of those, so treat them as the kind of thing to verify in your own account rather than a fact to plan around. Also in the same batch was a Landing page images option, which one write-up points out was already visible back in March, so the New label on it is a bit generous. Separately, Google has said image extraction switches on by default for newly created Performance Max campaigns from 4 December.

Here's why I think this one leads the issue. Until now your ad creative was a closed set, it was the things you made and uploaded, and if an ad was bad it was bad in a way you chose. This turns it into an open set, and the new members of that set are your own back catalogue, which almost nobody curates. A reel from two summers ago with a phone number in it you no longer answer. A video of a staff member who left on bad terms. A Christmas special from a year when you were cheaper. None of that was a problem while it sat quietly on your profile, because nobody scrolls that far back, and all of it becomes a problem the moment it's a paid ad being shown to strangers.

Do this week

Open each Performance Max campaign, go to Asset optimization, and find Sourced videos. Note whether it's on and which sources are ticked, because that tells you what the pool is. If you decide to leave it on, spend ten minutes scrolling to the oldest videos on every account you've connected and delete or archive anything with a dead price, a dead number or a person who no longer works there. If you'd rather not think about it, switch it off and keep uploading your own.

ChatGPT starts rewriting your ads on Thursday unless you tell it not to

OpenAI emailed ChatGPT Ads advertisers on Tuesday 20 October to say it is switching on something called creative enhancements, and the coverage landed on Friday 23 October across Search Engine Land, PPC Land, MediaPost, Shopifreaks and AdExchanger. What the setting does is let OpenAI rewrite your headlines and descriptions, translate your ad text, resize your images, and swap the destination URL for a different page on your own site. The versions it makes can run without your sign-off, and OpenAI's help text says there's no preview of every variation.

That last part is the actual change. Generated ad variations aren't new, advertisers had them before, but they had to be approved before they ran. This keeps the generation and drops the approval.

You wrote the headline and you chose the page, and from Thursday it can be a different headline pointing at a different page.

The dates don't agree, which is annoying and worth knowing. The email OpenAI sent says the feature is enabled starting 30 October. OpenAI's own help centre page says it turns on automatically starting Thursday 29 October, and one write-up notes the same page carries a line reading Thursday 31 October somewhere in it, which isn't a Thursday at all. I'm not going to pretend to resolve that, so the safe reading is the earliest of them: if you don't want this, deal with it before Thursday.

For an Australian reader this is live rather than theoretical, which is the correction we ran back in Issue 07. Ads inside ChatGPT reached Australia in late March, and businesses here have been able to buy them through agency partners or the self-serve beta since. Our advice then was to look rather than spend, and honestly that hasn't changed, because there are still no published small business benchmarks for the channel. But if you did open an account and leave a campaign running, this setting is now your problem rather than somebody else's.

The destination URL swap is the bit I'd actually lose sleep over. If you wrote an ad about one service and chose the page that sells it, and the system decides another page on your site is a better landing spot, then the thing you measured and the thing that happened are no longer the same thing. That's survivable on a big site with good pages everywhere. On a six page site where one page is good and the rest are from 2022, it isn't.

Do this week

If you have a ChatGPT Ads account at all, even a dormant one, open Ads Manager before Thursday and turn creative enhancements off at the account level. Opt-outs you already have carry over, and a campaign setting beats the account default, so check both. If you genuinely want to test it, turn it on for one campaign only and point that campaign at the one page on your site you'd be happy for anybody to land on.

Meta's ads got 14% dearer, nearly everywhere except around here

Meta reported its third quarter results on Wednesday 21 October, and the number advertisers care about is in there: the average price per ad was up 14% against the same quarter last year, while worldwide ad impressions grew 10%. Advertising revenue across the family of apps came in at 63.1 billion US dollars, up 26%. For context on the direction, price per ad was up 12% in the second quarter and impressions were up 14%, so prices are accelerating while the amount of inventory being added is slowing.

The regional split is the interesting part, and it comes from Meta's own earnings presentation, which allocates impressions and prices by where the user is rather than where the advertiser is.

Rest
+24%
US + Can
+23%
World
+14%
Europe
+10%
Asia-Pac
+1%

Change in Meta's average price per ad, third quarter 2026 against the same quarter last year, by the region the user is in. "Rest" is Meta's Rest of World, "US + Can" is United States and Canada, "World" is the worldwide average, and Australia sits inside Asia-Pacific. From Meta's third quarter 2026 earnings presentation, reported 21 October 2026. Impressions over the same period grew 14% in Asia-Pacific and 8% in each of the other three regions.

So Asia-Pacific had the fastest growth in the number of ads shown and almost no growth in what they cost, which is the opposite shape to everywhere else, and it's the second quarter running that Asia-Pacific pricing has been roughly flat. The plain reading is that more ad inventory is being created around here and there isn't much extra competition for it yet, whereas in the United States and Canada prices went up 23% on only 8% more inventory, which is competition doing exactly what competition does.

The 14% is the global auction talking, and the 1% is closer to the room we're actually standing in.

Now the honest limits, because a number this clean invites being misused. This is the entire auction, including the biggest advertisers on earth, averaged across Facebook, Instagram, Threads and WhatsApp, so it is not a benchmark for your account. Australia is one country inside an Asia-Pacific bucket that also contains India and Indonesia, and Meta doesn't break out countries, so the 1% is not an Australian figure, it's a figure that includes Australia. And what you personally pay is set by your audience, your industry and the time of year far more than by any regional average, which is why November and December are about to cost you more regardless of what this chart says.

What it's useful for is settling an argument, and the argument is usually some version of "everyone says Facebook ads have got too expensive." Globally, that's true and getting truer. In our part of the world it has been roughly flat for six months while the amount of ad space has grown faster than anywhere else, so if you've been putting off testing paid because you heard it's gone mad, the auction you'd actually be buying in is the calmest one on the list.

Do this week

Open Ads Manager, set the date range to the last 90 days, add CPM as a column, then compare it with the same 90 days a year ago. That's your price per ad, and it's the only version of this number that has anything to do with your budget. If yours is up a lot more than these figures, the cause is almost always your audience being too small or your creative being tired, not the market.

Most Australians are uneasy about what social ads show people doing

Roy Morgan published findings for Ad Standards on Thursday 22 October, picked up by Mi3 and Mumbrella the next day. The headline figure is that 87% of Australians think it's important that advertising shows safe and healthy behaviours, and that social media is the platform people are most uneasy about: 71% say they're worried about unsafe or unhealthy behaviour in advertising on social, which rises to 78% among people who use social several times a day. Underneath that, 63% are concerned about advertising that shows unsafe or unhealthy behaviour as fun, cool or desirable, and 61% are worried specifically about influencers showing unsafe behaviours.

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

The method, so you can weigh it: it's the Ad Standards and Roy Morgan Community Sentiment Tracker, run online with 1,032 Australians aged 18 and over between 17 August and 2 September. That's a reasonable national sample for a headline figure and a thin one once you start cutting it by age or state, so I'd take the shape and not the decimal places. It's also stated preference, which means it records what people say worries them rather than what they actually scroll past, and those two things are never identical. And it was in the field in August, so it isn't a reaction to anything that happened this month.

What makes it worth your time anyway is that it isn't really about advertising standards, it's about the ordinary stuff you'd film without thinking twice. Ad Standards is the body that handles complaints about Australian advertising, and the behaviours in that survey are not exotic. A tradie up a ladder with both hands full. Somebody taking a call while driving to a job. A hospitality reel where the drinks keep landing and nobody eats anything. A gym clip with form that would hurt a beginner who copied it. None of that is an attempt to be edgy, it's just what a real workday looks like when you film it, and a decent share of the country reads it as the ad endorsing the behaviour.

Nobody sets out to film something unsafe, they film a normal Tuesday and the normal Tuesday has somebody standing on the top rung.

The influencer number is the one I'd think hardest about, because 61% is a lot of unease about exactly the arrangement a small business is most likely to use, which is paying a local person with a few thousand followers three hundred dollars to make something. You have far less control over what they film than over what you film, and their video carries your name.

Do this week

Before the next piece of content with a person in it goes out, watch it once with the sound off and ask only one question: what is the person doing. Not what are they saying, what are they doing. If you're paying somebody else to make content, put the same question in writing before they film, because it's a thirty second conversation beforehand and a much longer one afterwards.

Meta pushed its new ad terms back to 20 November, and the fight is about your website

Meta said on Friday 23 October that it is moving the start date of its updated Commercial Terms and Self-Serve Ad Terms to 20 November, from 30 October. Reuters reported it and PPC Land, Relevant Audience and a long list of others followed. The reason is that publishers and digital advertisers complained to competition regulators in the European Union and the United Kingdom, arguing the new terms would hand Meta an unlimited right to use their content for its AI services. Meta rejects that reading of them.

Meta's own line, through a spokesperson, is that the extra time gives it more room to explain to partners and stakeholders what's changing and what's not. It says the revised terms apply only to its ads and business products, that it is not changing how it uses web crawlers, and that content owners can still use robots.txt to signal what they'd prefer. The European Commission said it welcomed the postponement and that its discussions with Meta are ongoing. Before the delay, Meta had already rewritten one sentence in the drafts, the one about content crawled from websites tied to business accounts being used to train its own products.

A caveat on the date itself, because it's a bit of a mess. When Relevant Audience opened Meta's own preview page on 24 October, the banner on it still said the terms take effect on 30 October, so Meta's page and Meta's statement to Reuters are currently saying different things. I'd watch for that banner to change rather than treat either date as settled.

Most of us will never read these terms, and that's exactly why it matters who pushed back on them and what they pushed back about.

Why I'm running this for a small business audience and not leaving it to the publishing trade press: if you run Meta ads, your business website is tied to a business account, so the question of what Meta may take from your site and what it may do with it is your question too. You are never going to negotiate these terms, nobody is, and there's no action here. What's worth having is the direction of travel, which is that the agreement sitting underneath your ad account is being rewritten to be more explicit about using your content in AI products, and the people who read it closely enough to object were the ones with lawyers.

The scope, for completeness, is four documents: the Commercial Terms, the Product Catalog Terms, the Self-Serve Ad Terms and the Ad Creative Generative AI Terms, with the AI creative terms being folded into the Self-Serve Ad Terms rather than sitting on their own.

Do this week

Nothing to sign and nothing to switch. If you want twenty useful minutes out of it, find out what your site's robots.txt currently says, which your web person can tell you in a sentence. Put 20 November in the calendar, and if your site carries writing or photography you'd mind being used to train something, have that conversation before then rather than after.

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 at the top, so whatever you used last week for writing and planning is still the right answer. The week's release was Claude Duet on Tuesday 21 October, a separate app where Claude runs as a background agent that checks your connected accounts and comes to you, built on Claude Opus 5.5. It's a research preview on a waitlist, for Pro and Max subscribers in the United States only, and Anthropic's own caveat is that it can still misjudge what's urgent. Still quoting no intelligence index score, seventh week running, because the public trackers disagree on what scale they're using.

Best for images

Still GPT Image 2.5, which is a repeat for the seventh week and we'd rather say so than dress it up. Nothing landed this week that changes it. Google's Nano Banana 2.1 is still the cheaper option if you're generating a lot or editing real photos rather than inventing scenes, and it's where we'd point a small budget.

Best for video

Something did finally move here, which is a change from the last six weeks, but we still can't name a leader honestly. Kling 4.0 went to a full global launch during October after a late September preview, and the public boards we can see disagree about where it and everything else sit, with different models on top depending on which board and which day. So the answer is the same as it has been: if you need video this month, film it.

Worth flagging

Look at what got automated this week. Google will find your videos, OpenAI will rewrite your copy, and the one genuinely new app is an assistant that watches your accounts and tells you what needs you. In all three the default moved the same way, which is that the software acts first and you review afterwards, if you review at all. That's a reasonable trade for some things and a bad one for others, but it only works at all if you know which defaults are on. So the habit worth building is boring: once a month, open every account you pay for and read its automatic settings. It takes twenty minutes and it is now the most valuable twenty minutes in the job.

Three platforms, one week, the same move

We went looking for five unrelated stories this week and kept finding the same one. Google switched on a setting that pulls videos out of your own accounts, OpenAI switched on a setting that rewrites your ad copy, and Meta's terms fight is about what a platform may take from your website. Three different companies, one direction, and in every case the thing that changed was a default rather than a feature.

We've started doing the monthly settings audit we keep recommending, because after Issue 10's Google promotions story we realised we were telling you to do something we weren't doing systematically ourselves. It takes about half an hour across the accounts we run, and the first pass turned up two settings we hadn't chosen. That's the whole case for it, really.

Still the honest caveat we run every week: our own numbers are thin while the case studies build out, so where we quote figures they're somebody else's and sourced, and where we're giving you an opinion we say it's an opinion. If you've found something in your own account this week that contradicts anything above, we'd genuinely like to hear it.