How marketing changes when buyers ask AI, and marketers work with agents.

• S1. From SEO to GEO. A 2024 research paper names GEO and tests methods on its own benchmark. It reports that adding citations, quotations and statistics raised visibility by up to 40 per cent in its tests, and that stuffing keywords gave little or no gain. That is one benchmark and a small set of engines, not a promise for any company. HubSpot’s recap of a talk by Aja Frost says to put the exact answer in the first sentence; that is a third party’s summary. L16 (from a third party’s summary; check before quoting) · L3 · Aggarwal and others, “GEO: Generative Engine Optimization”, KDD 2024 (arXiv)
• S2. From talking about ourselves to amplifying neutral voices. Kipp Bodnar’s published memo says brand mentions will matter more than backlinks and that individual voices do better than brand channels. Elena Verna writes that growth is now a trust problem. Frost’s recap says to earn mentions on sources AI tools already cite. These are opinions of the named people. Our own rule is firmer: we never create, buy or manufacture reviews, testimonials, analyst opinions or awards. L15 · L13 · L16 (from a third party’s summary; check before quoting)
• S3. From campaigns to always-on. Dave Gerhardt writes about doing short-term and long-term work together and testing a channel small before investing (P2), and about being consistent and patient (P8, a third party’s summary). Jay Acunzo says marketing is about who stays, not only who arrives. “Always-on” is our label; none of them is quoted as using it. P2 · P8 (from a third party’s summary; check before quoting) · L8
• S4. From a content calendar to a story spine. Emily Kramer says to define the few narratives you want the market to echo back, so you avoid random acts of marketing. Kipp Bodnar writes that if the strategy cannot be said in one sentence, the company is not ready to spend on it. “Story spine” is our term. The Past and Present-to-future format is borrowed from how a vendor keynote tells each story; it was read from an extract supplied by the owner, and none of that deck’s claims or numbers is used. L5 · L15
• S5. From gated assets to open assets. Latané Conant described an experiment of “no forms, no spam, no cold calls” and its reasons in her own talk. Dave Gerhardt says to give value before asking. That assistants can cite only what they can read is our inference, not a finding of those sources. Whether to keep any form is the CMO’s decision. L10 · P7
• S6. From attribution software alone to self-reported answers and simple signals. Chris Walker names these three methods in a LinkedIn post, and says attribution is for knowing which levers to pull, not for proving a channel. Dave Gerhardt makes the same point about channels that build up over time. This does not say that testing has no place for firms that can run it. L11 · P9 · P10 (from a third party’s summary; check before quoting)
• S7. From vanity metrics to share of answer. Rand Fishkin calls traffic a poor goal in a zero-click world and says to measure leads and conversion. Frost’s recap lists AI visibility, share of voice and citations as things to measure (a third party’s summary). The GEO paper defines visibility measures for generative answers. Share of Answer is a Foundronics measure; see the FOUND Standard for what it can and cannot say. L3 · L16 (from a third party’s summary; check before quoting) · Aggarwal and others, “GEO: Generative Engine Optimization”, KDD 2024 (arXiv)
• S8. From one-way outbound to community and partner-led. Kathleen Booth treats community as a long-term brand strategy with dedicated people. Elena Verna lists staffed community and matching creator partnerships among what builds trust. Partnership Leaders describe partner marketing as using each other’s strengths and audiences (a third party’s summary). Conant argues for relevance over volume. These are opinions and observations, not studies. L17 (from a third party’s summary; check before quoting) · L13 · PM1 (from a third party’s summary; check before quoting) · L10
• S9. From feature lists to the customer’s real alternatives. April Dunford starts positioning from the customer’s competitive alternatives and says to include the status quo. Only the free part of her second post was read. L1 · L2
• S10. From publish once to distribute where the audience already is. Ross Simmonds argues that distribution is a discipline and often matters more than the content itself (only the free chapters were read). Rand Fishkin says to be where the audience already is, even when no click comes back. L12 · L3

2. The claim register (template)

Before any asset is made, every claim and number goes in a register with these columns:

• ID
• Claim, in plain words
• Kind: fact, opinion or our claim
• Number, if any
• Source URL
• First-hand or secondhand
• Status: to be sourced, verified or retired
• Date checked
• Used in (letters and assets)
• Approved by

An asset that uses a claim marked “to be sourced” does not go out.

3. The content pack, mapped to the letters