PMI’s comms chief says SEO is over. She is half right.

Moira Gilchrist, chief global communications officer at Philip Morris International, told Business Insider’s CMO Insider series something that will be quoted in pitch decks for the next year: “The days of search engine optimization are over — now it’s all about generative engine optimization.” She is an early adopter by her own description, she has PMI’s corporate website being rebuilt for large language models rather than for search crawlers, and she has a second AI habit that is arguably the more interesting one — synthetic audience personas used to pressure-test messaging before it ships.

She is half right, and the half she is right about matters more than the half she is wrong about. The owned-channel argument is correct and most companies have not internalised it. The declaration that SEO is finished is the part that will cost people money, because generative engine optimisation is not a replacement discipline. It is the same discipline, pointed at a different consumer.

The owned-channel point is the real insight

Gilchrist’s observation about her own corporate site is the sharpest thing in the interview: “Nobody would go there, unless you were a journalist or perhaps an investor.” That was true of virtually every corporate website ever built. The company site was a compliance artefact and a press-release dumping ground, and the traffic numbers said so. Budget flowed to paid, to social, to the channels where the audience actually was.

Language models changed the economics of that page without changing its traffic. When a model answers a question about your company, your category, or your product, it assembles that answer from what it can retrieve — and your own domain is the highest-authority source it has for claims about you. The page no longer needs human visitors to do work. It needs to be retrievable, parseable, and unambiguous, because it is being read by a machine that will paraphrase it to someone who never visits.

This inverts a decade of content strategy. Pages that were written to rank — thin, keyword-padded, structured for a crawler counting terms — are actively bad inputs to a model that is trying to extract a factual claim. Pages that state things plainly, attribute them, date them, and structure them so a specific assertion can be lifted out intact are good inputs. The companies that treated their own site as a place to be precise rather than a place to be findable are, accidentally, ahead.

Why “SEO is over” is the wrong lesson

Strip the branding off generative engine optimisation and look at what the work actually is. You need pages a retrieval system can find and fetch. You need clean information architecture so the right page is the one that surfaces. You need structured data so entities and relationships are explicit. You need crawlability, sane canonicals, fast responses, and no JavaScript wall between the content and the fetcher. You need authoritative, well-sourced writing that other credible sites reference.

That is a description of technical SEO and editorial quality. Every single item on it was on the list in 2019. What changed is not the work but the scoreboard: you are no longer optimising for a ranked list of ten blue links, you are optimising to be the source a model reaches for and, ideally, cites. The tactics that were always cargo-cult — keyword density, exact-match anchor spam, doorway pages, content produced at volume to fill a topic cluster — die. The fundamentals get more valuable, not less, because a retrieval system is less forgiving of ambiguity than a ranking algorithm ever was.

There is a measurement trap here that catches people badly. Organic sessions will fall while your influence rises, because the model answers the question and the user never clicks. If you are still reporting on sessions, a successful GEO programme looks like a failure for two quarters. The teams we see handle this well have already moved their reporting to share of AI answers, citation frequency, branded search volume, and assisted conversions — this is the measurement rebuild that our SEO and growth work now starts with, because instrumenting the new scoreboard before you change the strategy is the only way to know whether the strategy worked.

The synthetic personas are the more interesting experiment

PMI built what Gilchrist calls “artificial audience replicas” to test messages before they go out. Her verdict: “Those personas are absolutely fascinating. The feedback that we get really helps us to improve how we’re communicating.” We have built versions of this for clients and the honest assessment is narrower than the enthusiasm suggests.

Synthetic personas are genuinely good at one thing: catching the message that lands badly. Ambiguity, unintended readings, tone that reads as defensive, a claim that invites the obvious hostile follow-up — a model roleplaying a sceptical regulator or an unconvinced customer will surface those in minutes, at a cost and speed no research panel can match. As a pre-flight check before you spend real money, it earns its keep.

What they cannot do is tell you what a real population thinks. A persona is a model’s compressed prior about a demographic, which means it reproduces the internet’s opinion of that group rather than the group’s opinion of itself, and it systematically under-represents the people least written about online. It will also agree with you more than a human would, because the assistant training that makes it useful also makes it accommodating. Used to replace audience research, it is a machine for confirming what you already believed, with a research-shaped artefact attached. Used to generate hypotheses that real research then tests, it is a real accelerant. The distinction is not academic — it determines whether the tool makes your messaging better or just makes you more confident.

The honest counter-argument

The strongest case against our position is that “it is just SEO fundamentals” is a comfortable thing for people who already do SEO to believe, and comfort is a bad signal. There is a real discontinuity buried in the continuity. Ranking was a competition among many results where being fourth still earned traffic; retrieval is closer to winner-take-all, because a model synthesising an answer pulls from a handful of sources and the rest get nothing. A long tail of pages that each earned a trickle of qualified traffic may simply stop working as a strategy, and no amount of technical hygiene fixes that — it is a structural change in how attention is allocated, and it favours a small number of canonical, heavily-referenced sources over a broad content estate. If that is where this lands, then telling clients to keep doing good SEO is technically true and strategically inadequate, and Gilchrist’s instinct to declare a clean break is the more useful provocation even if the literal claim is wrong. We think the fundamentals still carry most of the weight. We hold that view with less confidence than we held the equivalent view two years ago.

What to do in the next 30 days

• Ask the models what they say about you. Put your twenty most commercially important questions to the major assistants and record the answers and the citations. Most companies have never done this and are surprised, usually unpleasantly, by what comes back.
• Fix your own domain first. The pages making factual claims about your products, pricing, and company should be plain, dated, attributed, and structured. This is the cheapest intervention available and it is the one with the highest authority weighting.
• Check you are fetchable. Client-side-rendered content, aggressive bot blocking, and slow responses will quietly exclude you from the answer. Confirm what the retrieval agents actually receive, not what a browser renders.
• Change the report before you change the strategy. Add citation share and branded search to the dashboard now, so the coming decline in organic sessions is read correctly rather than triggering a panic.
• If you try synthetic personas, pair them with real research. Use them to find the failure modes, then validate with humans. Never let the persona be the last word.

Gilchrist is describing something real, and the companies that hear only the slogan will rebrand their SEO retainer as a GEO retainer and change nothing. The actual work is less exciting and more valuable: make your own domain the most precise, most retrievable, most citable source about your business, and measure whether models are picking it up. That is not the end of search engine optimisation. It is the first time in fifteen years that doing it properly has been more rewarding than gaming it.

Organic traffic falling while you have no idea whether models are citing you? Talk to Cravings about a generative engine optimisation audit — we measure what the assistants currently say about you, fix the retrievability of your own domain, and rebuild the reporting so you can tell progress from decline.