Your Next Audience Is A Machine: The Week In Marketing

The single most useful way to read this week’s marketing news is this: the machine is becoming the primary reader of your marketing, and the rules written for human readers are quietly being retired. OpenAI signalled that robots.txt may not govern one of its bots, Google pushed a faster AI model into Search results, and advertisers got a firm deadline to move off keyword-first campaigns. None of these were announced together, but they describe the same shift.

Underneath it is a simple change in the unit of marketing. For twenty years the unit was a human click. Increasingly, the unit is a machine interaction — a crawl, a retrieval, an AI-generated answer, an automated bid — and only some of those ever become a visit you can see.

1. The crawl bargain is being renegotiated

The open web ran on an unwritten deal: let search engines crawl your pages, and they send you traffic. Two stories this week showed that deal coming apart from both ends.

Search Engine Journal reported that OpenAI considers robots.txt — the plain text file publishers use to tell bots what they may access — not necessarily binding on its fetch bot, on the reasoning that a fetch triggered by a specific user request is different from a crawler indexing the web at scale. Whether or not you accept that distinction, it means the file you have used to control machine access no longer controls all machine access.

At the same time, Cloudflare’s projection, covered by Search Engine Journal, is that machine traffic could reach roughly a thousand times human traffic within five years. That is a hosting-cost story and a security story before it is a marketing story — some of that automated traffic is actively probing for credentials, not reading your blog.

Takeaway: stop treating robots.txt as a control and start treating it as a request. If you need real control over who reads your content, that now lives at the edge — server logs, firewall rules, rate limits and bot management, which usually means a conversation with whoever runs your infrastructure rather than your SEO agency.

2. AI answers just got cheap enough to be everywhere

Google began rolling Gemini 3.7 Flash into AI Mode in Search, as reported by Search Engine Land and Search Engine Journal. “Flash” models are the fast, low-cost tier of a model family. That detail matters more than the version number.

Expensive models get rationed to a small share of queries. Cheap, fast models get applied broadly. Every time the cost of generating an answer drops, the share of searches that receive a generated answer instead of ten blue links goes up. This is the mechanism by which AI answers spread — not a product announcement, but a unit-economics change.

Takeaway: assume the queries you currently rank for that are informational, definitional or comparison-shaped are on a slow path to being answered on the results page. Audit your top 50 landing pages and mark which ones survive that. The ones that do are usually pages carrying something a model cannot generate: original data, pricing, tools, logged-in experiences, local specifics.

3. The traffic that does arrive behaves differently

Search Engine Land published a piece on how LLM traffic converts differently — visitors arriving from large language model interfaces such as ChatGPT, Perplexity or Google’s AI Mode. The recurring pattern practitioners report is lower volume, later-stage intent, and a session that starts with the visitor already partly educated.

This breaks a lot of dashboards. If you judge that channel on sessions, it looks trivial. If you judge it on revenue per session, it can look excellent. Two other changes this week land in the same gap: Google Analytics added custom conversion attribution windows, and Search Engine Journal covered YouTube measuring creator video against branded search lift — an admission that last-click cannot see what a video actually did.

Takeaway: add two things to your reporting this month. First, a self-reported attribution field on your lead form or checkout — a plain “how did you hear about us?” box, which is now often more accurate than your analytics for AI-assisted discovery. Second, a branded-search trend line next to your spend chart. When your dark-funnel activity works, branded search moves before your attributed conversions do.

4. Ads stop asking what people typed

Google set a migration timeline for AI Max in Search campaigns, and Search Engine Land also argued that Demand Gen shows PPC can no longer hide behind intent. Read together, they say the same thing: the keyword is being demoted from a targeting instruction to a hint.

For a decade, paid search was the easy channel because someone else — the searcher — declared their intent, and you simply bid on it. As match types loosen and AI-driven campaign types expand reach, the platform decides who sees the ad. What you actually control shrinks to three things: your creative, your audience and conversion signals, and your landing experience.

Takeaway: treat the migration deadline as a forcing function on creative, not on settings. Before you migrate, make sure your conversion tracking distinguishes a good lead from any lead — send offline conversion or qualified-lead values back into the platform. An automated system optimising against a weak signal will confidently spend your budget on the wrong people, faster than a manual one ever could.

The India angle

Two of these threads bite harder in India. First, cost: if your D2C site or publication sits on a metered CDN or cloud plan, non-human traffic is a line item, not an abstraction. As a rough illustration, a brand paying ₹40,000 a month in bandwidth and origin costs sees that become a materially different number if automated traffic doubles — worth a look at your logs this quarter rather than next year.

Second, language. AI answers in Search are strongest at synthesising well-covered English content. Hindi, Tamil, Marathi and Bengali queries in specific commercial categories are still thinly served. That gap is a real, if temporary, opening for Indian marketers willing to publish genuinely useful vernacular content rather than machine-translated versions of their English pages.

What this means for you

Shift	Do this in the next 30 days
Bot access is no longer opt-out by file	Pull 30 days of server logs; identify your top non-human user agents and their bandwidth cost
AI answers spreading to more queries	Tag your top 50 pages as “answerable by a model” or “not”; reinvest in the second group
AI-referred traffic converts differently	Add a self-reported source field to forms; report revenue per session, not sessions
Keywords demoted in paid search	Feed qualified-lead values back to the platform before you migrate to AI Max

The connective idea worth carrying into your next planning meeting: you are no longer optimising a page for a person who will read it. You are optimising a source that a machine will read, compress, and repeat to a person you never see. Brand memory, original data and direct relationships are what survive that compression.

Frequently asked questions

Does robots.txt still work?

Robots.txt still works for the crawlers that choose to honour it, which includes major search engine crawlers. It does not reliably work for every AI-related bot — OpenAI has indicated that user-triggered fetches may fall outside it. Treat robots.txt as a stated preference and use server-level or CDN-level blocking when you need enforcement.

Will AI content be detected and penalised because of watermarking?

Not in the way most marketers fear. Search Engine Journal covered both Anthropic’s disclosure of how its watermark works and how it can be defeated and what such a watermark can and cannot tell you about authorship. A watermark can suggest text passed through a particular model; it cannot tell you who did the thinking, and normal editing weakens it. Search ranking still turns on usefulness, not on tool detection.

Is traffic from ChatGPT and AI Mode worth chasing?

Yes, but measure it on value rather than volume. Reports from practitioners consistently describe smaller numbers of visits arriving later in the buying process. Judge the channel on revenue per session and on assisted branded search, and expect it to under-report in last-click analytics.

What should a small marketing team prioritise from all this?

Three things, in order: get a self-reported attribution question onto your lead form, check your server logs for bot cost, and fix the conversion signal you send to ad platforms before automated campaign types spend against it. Everything else on this list can wait a quarter.