If ChatGPT referrals dipped in your analytics in mid-October, check the labels before you blame the channel.
AI visibility vendor Profound found that source tags in ChatGPT answers began disappearing on 8 October. By the evening of 14 October, only about 5% of the answers it collected that cited sources included a tagged link. By the morning of 16 October, that figure was back to roughly 99%.
Those percentages describe answers, not visitors. Referrer data can still identify some visits; missing tags alone don't tell you whether demand fell. A better-looking report needs the same scrutiny.
A business can become much better at running its marketing before it becomes better at growing.
More conversion credit doesn't mean more customers
Google's Campaign Type Attribution beta, announced earlier this year, lets Demand Gen assign conversion credit and bid separately from other Google campaign types. Google says account-level conversion totals on the overview and campaign pages will be blank while it's enabled, to avoid double counting.
The same conversion journey can receive different credit under different rules. The bidding system can then act on those rules.
That can help when comparing Demand Gen with paid social, provided you understand each platform's attribution scope. A more favourable allocation of credit doesn't mean the business acquired another customer.
Before changing the setting, save the reporting definition and comparison period. Record the switch date. Follow outcomes that keep the same meaning across it, such as enquiries that become customers or completed orders and their value. Investigate any gap between those records and the platform's improvement before increasing spend.
Google's Conversion Lift addresses the separate question of additional outcomes caused by advertising. It supports Search and Performance Max, but ask Google whether your account has access and enough data and budget for a useful study. For a smaller business, reliably connecting enquiries to sales may be the more urgent job.
The IPA's 7 October promotion research exposed a more basic problem. Among 250 senior UK consumer-goods marketing and insight leaders surveyed by Censuswide, 75% couldn't estimate what proportion of their previous year's promotions had been profitable. Yet 82% believed their latest promotion was profitable.
Confidence was considerably easier to find than a reliable estimate.
Start with the activity the business keeps repeating. If its profitability is still a guess, another reporting view won't settle the question.
Google added controls. Braze automated another decision.
Google's channel prioritisation and budget-allocation controls began a phased rollout from 15 October. Advertisers could express preferences across eligible channel groups and set certain spending limits.
Ordinary allocation percentages are best-effort limits against average daily budget. A supported 0% allocation prevents spend on that channel. Demand Gen gets prioritisation, but not budget allocation.
Prioritisation has no effect when a campaign isn't limited by budget; supported allocation limits still apply. Ask whoever runs your Google Ads which setting the campaign needs. A preference and an exclusion are different decisions.
Braze's Decisioning Studio Go became generally available for Braze customers on 14 October, initially optimising email call-to-action clicks within marketer-defined options and limits. Six days later, Salesforce agreed to acquire Braze for about $5.66bn in enterprise value, with completion expected in Salesforce's fiscal 2028, subject to approvals and other conditions.
If you're choosing a lifecycle platform, judge what it can do today. Braze said actual integration would begin only after closing, although planning could begin beforehand.
For the live feature, ask what those email clicks should help someone accomplish. If they lead to a confusing offer, better email optimisation can deliver more people to the same unresolved problem.
Removing manual work has value of its own. Count the time or cost saved, then assess any claim about additional demand separately.
ChatGPT ads: the investment question behind the advertiser count
On 21 October, OpenAI said more than 100,000 advertisers had run ChatGPT campaigns. Expansion into 26 additional countries brought availability to 89; UK availability had already been confirmed in August. Ads reach eligible Free and Go users. Plus, Pro, Business and Enterprise users don't see them.
The commercial possibility is a placement while someone explores needs and compares options in conversation. The advertiser count doesn't tell you how many kept spending, what customers they acquired or what another advertiser should expect to earn.
I wouldn't move money from a working channel on the strength of that count. If the audience and offer fit, run a capped test with an agreed outcome and a reason to stop.
Google announced AI Mode ads in ten more markets on 28 October. The UK remained a Q4 plan at month-end. Check actual availability and reporting before putting it into a November media plan.
Your website's offer can become your ad's promise
Google told some advertisers by email that from 12 October it would extract website offers and turn them into promotion assets. Its help page specifies eligible Search and Performance Max campaigns with location assets and no manual promotion running. Advertisers can switch the feature off or supply promotions themselves.
That makes the landing page part of ad production. Whoever edits the offer on your website may also be writing part of your ads.
This is the change I'd prioritise for November: review the source offer and its conditions before the advertising system reuses it. Reviewing the finished ad alone can miss where the promise began.
On 14 October, the CMA fined Wayfair £4.8m over false time-limited offers. Its countdown deals created pressure to act, yet many products remained at the same price or cheaper after the deadline. Appliances Direct was fined £1.2m for misleading time limits and automatically adding paid extras. Gold's Gym (UK) Limited, an independent franchise operator, was fined £45,000 over unclear or late presentation of membership prices, including joining fees. All three admitted breaches and settled.
On 28 October, the ASA upheld a complaint about an Attraction Tickets Google ad, seen in August, promising: “Find A Lower Price & We'll Match It”. The guarantee only covered competitors belonging to ABTA or ABTOT. The ASA said that significant restriction should have appeared in the ad itself. Being at least two clicks away wasn't enough.
These cases concern the promises themselves; they aren't evidence that the companies used automated promotions. For anyone adopting offer extraction, they give the source-page review a clear purpose.
Sometimes the strongest conversion work starts by rewriting the promise.
Consumer confidence can't diagnose your pipeline
The Q3 IPA Bellwether report, published on 15 October, found that 25.6% of respondents had raised marketing budgets and 11.3% had cut them. The +14.3% net balance measures the direction of revisions, rather than growth in pounds spent.
Then GfK's Consumer Confidence Barometer reported an 11-point fall to −24. Interviews ran from 1–14 October, before the UK Budget on 28 October, and measured attitudes rather than purchases.
If a finance director cites falling confidence as a reason to cut spend, bring the discussion back to your customers. Has qualified demand weakened? Are decisions taking longer? What are people saying when they stop? Equally, a wider rise in marketing budgets is a poor reason to increase yours without a spending case.
An affordability problem, a lack of urgency and uncertainty about which option to choose call for different responses. Defaulting to a discount avoids making that distinction.
Help the buyer who doesn't know how to choose
Sephora's 26 October gifting campaign announcement described Seth Rogen choosing presents with help from a Beauty Advisor, alongside a curated selection for the US and Canada campaign.
The interesting choice is who gets the expertise. A gift buyer may know far less about beauty than the person receiving the gift. They can be willing to spend and still feel poorly equipped to choose.
That distinction also matters when someone commissions a service or approves software that another team will use.
More product claims may leave that hesitation untouched. Guidance has a clearer job: help the buyer reach a suitable choice. The announcement reported no sales results, but the design gives a specialist business something worth testing.
Could someone unfamiliar with your category choose well without learning all your terminology first? I'd use that question to brief the next piece of creative.
The November watch: Meta's website-content terms
The CMA's 30 October investigation concerns terms for businesses and publishers using Meta's services. They could let Meta collect website content for its own products, including AI features, without those businesses choosing how it is used. Meta paused the rollout on 28 October, superseding the earlier 20 November implementation date. No infringement finding had been made. If your business uses Meta's services, read the revised terms when they appear.
Before the next budget increase
| Before you… | Ask for |
|---|---|
| Believe an improved conversion report | The attribution-change date, consistently counted orders or converted enquiries, plus what those customers are worth after discounts, returns and the costs of winning and serving them. |
| Change channel controls | The setting that actually excludes spend, distinguished from a preference or best-effort limit. |
| Distribute an offer automatically | The source page, ad and qualifying terms side by side, checked as a first-time buyer. |
| Fund a new-channel test | The intended customer, a capped budget and the outcome that determines whether spending continues. |
By 16 October, the tags in Profound's sample had largely returned. Their recovery still didn't tell a business how demand had changed during the interruption.
When a number moves, find out what moved with it. Check the measurement, then follow the customer far enough to see what happened.
Before you call it growth, know which explanation you can support.
A selected review of October 2026, with sources linked throughout and key claims checked on 3 November. Reporting cut-off: 31 October.
