The dials are gone: what mattered in marketing this week

Four stories from this week point in the same direction: the levers marketers used to pull by hand — match types, placements, audience segments, even the click itself — are being handed to machines, and what is left under your control is the quality of what you feed them. Google is expanding AI Overviews on the fly, a new open standard would let AI agents take actions inside your website, and marketers are openly admitting their AI tools are running on bad data.

Put simply: the job is shifting from operating the dials to auditing the inputs. Here are the four threads worth your attention, and what to do about each one.

1. Google’s results page is now a moving target

Search Engine Land reported that Google is dynamically expanding AI Overviews for some queries — meaning the AI-generated answer block at the top of results can grow or shrink depending on the query, rather than appearing in one fixed shape. AI Overviews, for anyone new to the term, are Google’s machine-written summaries that sit above the traditional blue links.

That matters because it breaks a habit most SEO (search engine optimisation) teams still have: treating a keyword’s search-results layout as stable. It isn’t. The same query can show a short answer today and a sprawling one next month, and your click-through rate moves with it even when your ranking doesn’t.

Alongside that, Search Engine Journal’s SEO Pulse roundup covered continuing fallout from Google’s latest spam update and questions raised by the judge in the US search antitrust case about Google’s AI products. The regulatory thread is slow-moving, but it is worth tracking: the remedies shape what Google can do with AI in search.

Takeaway: stop reporting rankings on their own. Report rankings alongside impressions and clicks from Google Search Console, so you can see the gap open up when an AI Overview eats your click. If you publish news or original reporting, Search Engine Journal also has a practical walkthrough for adding Google’s Preferred Sources button to your site — a small piece of code that lets loyal readers tell Google they want to see more of you. Availability varies by region, so check before you promise it to a client.

2. Your website is about to get non-human users

The most under-covered story of the week: WebMCP, a proposed standard that lets AI agents perform actions inside websites — not just read a page, but use it. Think searching a catalogue, filtering results, or starting a booking, driven by an assistant rather than a person with a mouse.

Pair that with MarTech’s report that an Anthropic partnership makes Salesforce’s interface optional, and the pattern is clear. Software is being unbundled from its screens. If a sales rep can ask an assistant to update a pipeline instead of clicking through a CRM (customer relationship management system), the interface stops being where the value lives.

For marketers, the uncomfortable implication is that your beautifully designed landing page may increasingly be consumed by something that cannot see design at all. Structured data, clean HTML headings, honest product feeds and machine-readable pricing become conversion assets, not technical housekeeping.

Takeaway: audit one high-value page this month for machine readability. Does it state the price, availability, location and key specification in text rather than inside an image? For an Indian D2C brand, that includes obvious things a bot cannot infer — whether the ₹2,499 price includes GST, and which pin codes you actually deliver to.

3. AI didn’t break marketing accountability — it read it out loud

Two pieces landed on the same nerve. MarTech reported that marketers know AI is using bad data to make decisions, and Search Engine Journal argued that AI isn’t killing marketing accountability, it’s exposing who never had it.

Both point at the same mechanism. An automated bidding system optimises towards whatever you told it success looks like. If your conversion event fires on every form fill including the spam ones, the machine will efficiently buy you spam — faster and at greater scale than any human media buyer ever could.

Search Engine Land’s piece on whether you want more leads or better leads is the practical version of the same argument: the choice is made in your conversion setup, not in a strategy deck. And MarTech’s argument to stop measuring your brand and start listening is a useful corrective for teams whose brand tracking has become a quarterly number nobody acts on.

Takeaway: spend an hour on your conversion definitions before you spend another rupee on automation. Feed qualified-lead or revenue signals back into the ad platform, not raw form fills. On the tooling side, Search Engine Land’s framing is the best rule going: “Use Claude for SEO. Don’t let Claude do SEO.” The same applies to any assistant — useful for clustering, drafting and analysis, dangerous when it publishes unsupervised. Their companion piece on seven ways to use AI for SEO work that matters is a sensible starting list.

4. When targeting automates, creative becomes the lever

Google told advertisers that better YouTube creative can more than double return on investment. Treat that as vendor-supplied research rather than independent proof — Google benefits from you making more video. But the direction is consistent with what practitioners see. Greg Finn’s reflection on 20 years of PPC, from exact match keywords to AI Mode, describes the same arc: as the platform absorbs targeting decisions, the inputs you still own — creative, offer, audience data, landing experience — carry more of the outcome.

The offline version of this appeared at Old Navy, which is rewiring its marketing strategy after a summer traffic bust, per Marketing Dive. No amount of media optimisation rescues a proposition customers have stopped walking in for.

Takeaway: shift budget from campaign micro-management to creative volume. If you run ₹5 lakh a month on paid media, a fortnight of an analyst’s tinkering is worth less than four new video variants and two new offers tested honestly.

What this means for you

• Rebuild your search report. Add impressions-to-clicks ratio by query group, so AI Overview erosion shows up as a trend rather than a surprise.
• Fix the conversion event first. Automation multiplies your definition of success. A wrong definition scales the waste.
• Make one page agent-readable. Prices, availability, delivery areas and specifications in text. Structured data where it applies.
• Budget for creative, not for fiddling. Set a monthly quota of new creative variants and protect it.
• Keep a human in the publish step. Use AI for research, clustering and first drafts; keep approval with someone accountable for the outcome.
• Watch the plumbing. If you buy programmatic media, note Google’s October changes to the Display & Video 360 API and check with your agency who owns the migration.

Frequently asked questions

What are AI Overviews and why do they affect my traffic?

AI Overviews are Google’s AI-generated summaries shown above the standard search results. When one appears, the traditional links move down the page and many users get their answer without clicking. Search Engine Land reports Google is now expanding these summaries dynamically for some queries, so the amount of space they take can change over time for the same search.

What is WebMCP?

WebMCP is a proposed standard that lets AI agents take actions inside a website — such as searching, filtering or submitting a form — rather than only reading its text. If adopted widely, it means websites will need to be usable by software agents, not just by people looking at a screen.

Should I let an AI tool run my SEO?

No. The practical consensus among search practitioners is to use AI assistants for research, keyword clustering, data analysis and first drafts, while keeping strategy, fact-checking and publishing decisions with a human. Unsupervised AI publishing is exactly the pattern Google’s spam updates target.

What is the single most useful thing to fix this week?

Your conversion definitions. Automated bidding optimises towards whatever event you mark as success, so if unqualified form fills count as conversions, the system will buy more of them. Feeding back a qualified-lead or revenue signal is usually the highest-leverage change available to a small team.