Nicola Sirchia

Short notes and long reads

Grouped by topic. Every long read takes five minutes or less.

Agent-ready brands

How a brand gets found, understood and chosen when an AI assistant builds the shortlist.

4 short notes

The ghost brand

A brand can be on the shelf, with loyal buyers and a fine media plan, and still be missing when someone asks an assistant for a recommendation. It rarely comes from a bad product. It comes from incomplete product data, claims a machine can't check, and content made for a campaign instead of a question.

Who owns share of answer?

Gartner expects 70% of enterprise brands to report share of answer to leadership by 2027. The brand team gets the new number, but the inputs sit with e-commerce, regulatory, customer service and sales. Name one owner for the inputs before the first report.

When most searches end without a click

SparkToro's research finds that 68% of Google searches now end without a click to another website. Best practice used to mean optimising for an algorithm. Now it means building a brand people look for by name, and being the source an AI answer cites.

A cart that follows the shopper

When an AI-powered cart follows people across search, video and email, brands compete for interpretation as well as attention. Can an assistant understand the product, compare its value and trust its claims?

Trust and governance

Who is responsible when AI acts in a brand's name, and what that asks of marketing.

5 short notes

Who pays for a platform's AI mistakes?

When an ad platform's AI can rewrite copy and add formats to an ad you already approved, and its terms leave the responsibility with the advertiser, that is a governance decision. Ask which AI features are switched on, decide which categories get no automatic edits, and name one person who checks the ads as they actually run.

How your AI treats a difficult customer is part of the brand

A brand agent meets rude customers every day. What it answers, and when it hands over to a person, usually ends up as a default setting chosen by a developer. It deserves the same care as the logo.

Recommendation integrity

Microsoft has described AI recommendation poisoning: hidden instructions in links and summarise buttons that shape what an assistant remembers and recommends. The next battleground after search ranking is whether a recommendation was earned or manipulated.

When the customer becomes a supplier

A patent application describes training AI on the idle graphics cards in people's homes, in exchange for credits. Very few people would agree to that with a brand they don't already trust. They would expect a clear explanation and an opt-out that is easier to find than the opt-in.

The microsite nobody owns

Every system has a business owner and a technical owner. The microsite from a launch three years ago has neither, and that is usually the real security risk. Keep one list of every site, shop and landing page you own, including the ones agencies host.

Building with AI

What I learn when I build something myself.

1 short note

An advisor people can trust with a real repair

It started in front of a shelf, watching a man pick up one product after another without knowing which would do the job. A chatbot is quick to build. An advisor people can trust is a different story, and the hardest parts are the ones nobody sees in a demo: safety guardrails, the handover to a person, and a rating on every conversation.