Presence, Not Channels

The problem is not how much companies spend on AI, but that the properties they own were never built to earn presence inside answer engines.

By Stefano Zoia · · 6 min

Somewhere this quarter, a service director is defending a chatbot roadmap to a board. The slides are good. Containment is up, handle time is down, the curves all bend the right way. Nothing in the deck is false.

Here is what the deck does not show. In their most recent service interaction, customers were roughly three times more likely to open ChatGPT, Gemini or Copilot than to open that chatbot. Use of third-party assistants has nearly doubled in a year. Use of company-provided chatbots has not moved since 2022.

The tempting conclusion is that the money is in the wrong place. It is not. The money is buying the wrong kind of thing. Almost everything a company owns online was built to be browsed by a person, and almost none of it was built to be read, trusted and repeated by a model. No amount of additional spend on the same category of asset closes that gap.

Where the word “channel” came from

Channel descends from the Latin canalis, a pipe or conduit. A canal is something you dig, line and maintain, and its entire promise is that water goes where the digger intended. The metaphor arrived in commerce through distribution — physical goods moving down a route you controlled — and it held for as long as the route was the scarce thing.

Digital marketing inherited the word and the assumption underneath it. A channel was a place you built, measured and owned: the site, the app, the newsletter, the branded assistant. Strategy meant choosing which conduits to dig and how wide. Every one of those conduits was shaped around a person arriving, looking and clicking.

That shape is now the expensive part. When the customer’s first move is to ask a general-purpose assistant, there is no conduit to dig. There is only the question of whether your information is structured, sourced and authoritative enough for the assistant to carry into its answer. Channels are built for arrival. Presence is built for retrieval. They are not the same asset, and they do not respond to the same investment.

Three findings, one shape

The wrong property. Service and support leaders put a median of twelve per cent of their 2025 budget into AI, the highest share of any function assessed, and roughly a quarter of them could show a positive financial return. It is worth being precise about why. A conversational layer bolted onto a website does not make the information underneath it any more legible to a model. It is one more human interface, competing against assistants that hold more context and recall it better. The spend is not too small. It is aimed at the wrong property.

The extraction gap. A three-year Press Gazette analysis of monthly visits, published this month, dates the traffic collapse to a July 2024 peak: the fifty largest United States news sites fell by more than a third, and generative AI referrals now account for a fraction of one per cent of what arrives. TollBit’s European data is starker, at roughly one human referral for every 179 AI bot visits, about three times worse than North America, with the ratio widening sharply between the first and second quarters of this year. Machines are reading company and publisher material at industrial scale and sending back almost nobody. Whatever is being extracted, it is being extracted from properties that were never designed to be quoted.

The quality inversion. And yet the few who do arrive are worth more. Adobe’s July cut, published on 19 August, has AI-referred retail visitors converting sixty per cent higher than non-AI traffic, up from fifty-four per cent three months earlier, with revenue per visit more than half again as high. Fewer visitors, better visitors. That is what turns this from a defensive problem into an investment case: presence in an answer has a measurable return, which means the asset that produces it can be funded like any other asset.

These three findings do not agree about whether the situation is good or bad. They agree about what is failing, and it is not the budget line.

What a property built for recall looks like

The honest reason the wrong asset keeps getting funded is that ownership is easy to measure and legibility is not.

Gartner’s August releases make the pattern almost comic in its consistency across functions. Ninety-three per cent of audit leaders report using AI; thirty-eight per cent have a strategy for it. Sixty-seven per cent of supply chain digital investment now goes to AI; fifty-five per cent of chief supply chain officers cannot say what it returned. Different functions, different instruments, identical shape. Organisations are counting activity and reporting it as transformation.

The adoption dashboard is the name I would give this. It is an instrument that is accurate about the wrong quantity. It tells you how much AI you have bought. It cannot tell you whether a machine that read about your category in Zurich or Milan this morning came away able to describe you correctly, and that second question is the one that decides whether you appear in the answer at all.

What replaces it is unglamorous and specific. Content modelled as structured, atomic units rather than monolithic pages, so a claim can be lifted without dragging a layout with it. Schema and semantic metadata produced as a default output of the content system, not retrofitted by an agency once a year. Product and service information held in one place that every surface can reach, rather than trapped in the page it happened to be written for — the compatibility fact, the warranty term, the dimension, the care instruction. Question-and-answer content types written to be extracted rather than browsed. And underneath all of it, first-party authority: claims that carry a source, a date and a name, because a model choosing between two accounts of your product will take the one it can attribute.

What presence actually costs

Presence is not free, and the two most useful August findings on that point pull in opposite directions.

Gartner’s inference forecast, published on 17 August, argues that cost per agentic workflow will rise more than fivefold through 2028, because falling token prices are quietly subsidising ever more complex work. Unit economics improve while the bill climbs. Anyone budgeting on the assumption that models get cheaper is budgeting for the wrong curve.

Against that, the same firm’s guidance to finance leaders three days later argues that the first agent should be run as a governance pilot rather than a return-on-investment pilot, because early agent programmes fail on unclear controls rather than on technology. Boundaries — what the agent may read, what it may do, where a human must sign — are set before development or not at all. A pilot is judged on one thing: whether you can reconstruct what the agent planned, accessed and produced.

The Swiss and Italian markets I work in are unusually exposed to all of this, because a great deal of commercial authority in both still lives in relationships and in documents written for humans. Neither travels well into a model. The firm with thirty years of earned trust and a website full of PDFs is, to an answer engine, quieter than a competitor half its age with clean structured data.

The channel was a conduit you dug. Presence is a reputation you maintain inside systems you cannot instrument. Which leaves the question I have deliberately not answered here: what becomes of the assistant you put on your own property. Is a first-party agent the last of the old channels, or the first of the new properties — and what would have to be true underneath it for the second answer to hold? That is the next piece of work.

Stay close.

— Stefano

Sources

  1. Gartner, Gartner Survey Finds Customers Are 3x More Likely to Use Third-Party GenAI Than Company-Provided Chatbots for Customer Service, 8 July 2026. https://www.gartner.com/en/newsroom/press-releases/2026-07-08-gartner-survey-finds-customers-are-three-times-more-likely-to-use-third-party-genai-than-company-provided-chatbots-for-customer-service
  2. Gartner, Gartner Survey Finds 87% of Customers Say Companies Using GenAI for Customer Service Must Provide Access to a Human Agent, 4 August 2026. https://www.gartner.com/en/newsroom/press-releases/2026-08-04-gartner-survey-finds-87-percent-of-customers-say-companies-using-genai-for-customer-service-must-provide-access-to-a-human-agent0
  3. Press Gazette, Traffic to 10 Biggest US News Websites Down a Third in Two Years, 21 August 2026 (Press Gazette analysis of monthly visits; the figures cited above are the top-50 numbers reported in the body, not the top-10 headline figure; United States only). https://pressgazette.co.uk/media-audience-and-business-data/media_metrics/traffic-to-10-biggest-us-news-websites-down-a-third-in-two-years/
  4. Digiday, European Publishers Are Getting Hit Harder by AI Bot Scraping, Report Finds, 14 August 2026 (TollBit data). https://digiday.com/media/european-publishers-are-getting-hit-harder-by-ai-bot-scraping-report-finds/
  5. Adobe Digital Insights, AI Traffic Trends Report: August 2026, 19 August 2026 (United States retail and travel data). https://business.adobe.com/resources/sdk/ai-traffic-trends-report-august-2026.html
  6. Gartner, Gartner Survey Finds Audit Teams’ AI Use is Common, But Most Teams are Lacking Strategic Adoption and Application, 11 August 2026. https://www.gartner.com/en/newsroom/press-releases/2026-08-10-gartner-survey-finds-audit-teams-ai-use-is-common-but-most
  7. Gartner, Gartner Survey Finds Majority of Chief Supply Chain Officers Unclear on AI Investment Returns, 5 August 2026. https://www.gartner.com/en/newsroom/press-releases/2026-08-05-gartner-survey-finds-majority-of-chief-supply-chain-officers-unclear-on-ai-investment-returns
  8. Gartner, Gartner Predicts AI Inference Costs Per Agentic Workflow Will Increase More Than Fivefold Through 2028, 17 August 2026. https://www.gartner.com/en/newsroom/press-releases/2026-08-17-gartner-predicts-ai-inference-costs-per-agentic-workflow-will-increase-more-than-fivefold-through-2028
  9. Gartner, Gartner Says CFOs Must Pilot Governance First Before Scaling AI Agents, 20 August 2026. https://www.gartner.com/en/newsroom/new/pr-q-a-new-vis-template/2026-08-20-gartner-says-cfos-mus-pilot-governance-first-before-scaling-ai-agents
  10. Harvard Business Review, Paul F. Accornero, Algorithmic Shopping Is Here. Is Your Company Ready?, 7 August 2026. https://hbr.org/2026/08/algorithmic-shopping-is-here-is-your-company-ready
  11. McKinsey & Company, Rewiring Advertising: The AI Imperative for Commerce Media Networks, 5 August 2026. https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/next-in-growth/rewiring-advertising-the-ai-imperative-for-commerce-media-networks