CMS Summit 26 Frankfurt: A Write-Up

I have just returned from CMS Summit 26 in Frankfurt, held at the Museum für Kommunikation on 12 and 13 May, and as is my habit after these events, I want to put down some thoughts while it is all still fresh.

It was a good conference. I was a speaker on the Tuesday morning, sharing the stage with people whose work I have followed for years and people I had not met before but will now be keeping an eye on. The shape of the programme leaned heavily into AI and LLMs, which is exactly where the CMS conversation needs to be in 2026, and Janus Boye, as ever, had curated it so the talks reinforced each other rather than overlapping.

Thanks first

Janus Boye hosted. Anyone who has attended a Boye event knows the rhythm: short talks, real participation, roundtables that actually move, none of the dead air that turns most conferences into a slog. CMS Summit 26 was no exception. The two days felt full without feeling forced, which is the hardest balance to strike.

Matt Garrepy was MC. Matt opened both days, ran the European CMS Idol competition on Tuesday evening, and held the room throughout with the kind of light-handed authority that lets a conference breathe. He also gave the Tuesday morning "what is now and what is next" analyst slot, which set up the rest of the programme cleanly. Having someone who can both moderate and contribute is rare; Matt does both well.

And Dr. Corinna Engel from the Museum für Kommunikation Frankfurt deserves a mention for welcoming us into a venue that genuinely fitted the theme. A communications museum hosting a conversation about machines learning to communicate, the symmetry was not lost on anyone.

Tuesday, the day

After Matt's opener, Kate Kenyon (JPMorganChase) delivered what was, for me, one of the two standout talks of the conference: The hidden work behind AI-ready content. Kate's argument is that companies have invested heavily in editorial and UX writing while underinvesting in the architecture, models, and processes that make content usable by both humans and machines, and the gap shows up sharply when AI enters the picture. She drew on operating one CMS for millions of customers at scale, with honest examples of what works, what does not, and what teams consistently underestimate. The candour landed. There was no vendor pitch, no slide of best practices that nobody will follow; just a working practitioner showing the texture of the problem. I took more notes during Kate's session than any other.

Florian Keitgen (b13) followed with The hidden scaling risk: why the human layer breaks your organisation, using open-source communities as a lens on collaboration dynamics. His point, that informal authority, recognition, and communication patterns weaken company performance long before technology does, is one that anyone running a digital programme will know. It is also one most conferences avoid because it does not have a tool to sell. Florian's willingness to sit with the discomfort of the problem rather than rush to a solution made it stronger.

After coffee, Stine Ferse (DHL) gave the practitioner's view from inside a global content platform: Running a global content platform: lessons from the real world. Stine spoke about governance, cross-business-unit collaboration, and the unglamorous infrastructure of making a CMS work at global scale. The honesty about what proved harder than expected was the part that resonated; every large-scale CMS programme has these stories, and few speakers will tell them publicly.

My own slot followed: The Web Has a New Audience. I will not write up my own talk except to say that the questions afterwards were sharp and useful, and several conversations carried on into lunch in a way that suggests the framing landed.

Liz Nelson (Sitecore) closed the morning with The Internet Talks Back. The title alone gives you the spirit. Liz's vantage point as VP of Product and Technology at Sitecore is unusual; she sees what enterprise CMS customers are actually asking for, and what they will not adopt, and her remarks reflected both.

The afternoon roundtables were a highlight in their own right. Nicole Rogers ran AI agents, Kate Kenyon ran content design, Antonia Fedder ran digital accessibility, Jeffrey McGuire ran digital sovereignty, and I ran a table on MX which generated more discussion than I had room for. The Boye roundtable format, small tables, twenty-five minutes, switch, remains the most efficient way I know to actually exchange knowledge with peers.

After coffee, Jeroen Fürst (TrueLime) gave The End of Platform Lock-In? Vibe Coding, MCP, and What Agencies Must Become, a sharp piece on how Model Context Protocol and AI-generated implementation are eroding platform-specific expertise as a differentiator. Jeroen's point that "content models, governance, stakeholder alignment, and platform judgment now determine whether systems hold together or fall apart" is the right one. Agencies that build only on platform-specific expertise should listen to this talk on repeat.

Chad Solomonson (RDA) followed with Build Smarter, Ship Faster: The AI-Composable Roadmap. Chad's framing, AI embedded into composable architecture rather than bolted on, is the right one for enterprise teams currently trying to figure out where in their stack AI actually belongs. Practical, with concrete delivery patterns.

Antonia Fedder closed the Tuesday talks with Where digital bias hides: configuration, content, and communication. Antonia's session moved between three layers, configuration, content, and communication, showing how everyday decisions encode assumptions about who the user is, and who gets left out. Her framing of bias as something that "reaches every person who interacts with your product" rather than something abstract was the move that made the session land. I left with a set of questions I will use on my next project.

European CMS Idol 2026 followed, hosted by Matt, with CKEditor, Griddo, Neos, TYPO3, and Webiny each given six minutes to make their case. Markus Schork, Antonia Fedder, and Matt McQueeny judged. I will not spoil the result for anyone who has not seen the announcement, but the format, six minutes, no slides of cruft, expert panel commentary, is the right one for a vendor showcase, and I hope other conferences copy it.

The evening dinner at Apfelweinwirtschaft Frau Rauscher rounded out the day. Frankfurt apple wine, good conversation, and the kind of unstructured time that turns conference acquaintances into actual contacts.

And then, after the food, Matt Garrepy's Elvis impersonation came as a complete surprise. Well done, Matt. I do not know how you do it, but it was the right note to end the day on.

Wednesday, the day

Nicole Rogers (ai12z) opened the second day with How AI is Reshaping Discovery, Websites, and Personalization. Nicole's session named AEO, GEO, and AIO directly, and her vantage point as a co-founder building in this space gave the talk a sharpness it would not otherwise have had. More on this in the side note below.

Søren Schaffstein (dkd) followed with Precious Users, Turning Fewer Clicks into Higher Conversions. The premise, that fewer clicks but higher intent is the new shape of web traffic, is correct, and Søren's practical framework for handling the shift was usable rather than aspirational. A talk for anyone whose dashboard is showing declining sessions and rising conversion rates at the same time.

Jeffrey "jam" McGuire (Open Strategy Partners) then gave the second standout of the conference for me: Beyond Burnout and Buyouts: A Third Way for Open Source CMS. jam's argument is that the funding model that built open source is breaking, contribution is at historic lows, the EU Cyber Resilience Act is about to make every implementer legally liable, and the most "successful" open source exits of the last decade ended in private-equity extraction. His proposal of community-owned commercial stewardship as a third way between volunteerism and venture capital was the most substantive piece of new thinking I heard at the conference. It is the kind of talk that will be quoted and argued with for the next year, which is the best thing one can say about a session.

After coffee, Markus Schork (Codal) gave WYMIWYG: What You Model Is What You Get, a clean dissection of why editing a website is still so hard decades into the CMS era. His point that frontend components should not dictate content structures, and that many CMS fields are only ever changed once, is the kind of observation that comes only from working with real systems for years. The "flexibility trap" framing is one I will be using.

Ondrej Chrastina (CKEditor) closed the morning with The real story behind AI in content editing, an honest look at where enterprise AI-in-content actually stands today, drawing on discovery work and customer conversations rather than vendor optimism. The candid pattern he described, fragmented "bring your own AI" workflows lacking context and governance, matches what I see in client work, and naming it publicly is useful.

Matt McQueeny (iMedia) gave the closing keynote, From Backlog to Boardroom: Fewer Clicks, Higher Stakes. Matt's thesis, that AI visibility has moved from a marketing or technical issue to a boardroom conversation, with direct implications for traffic, revenue, and competitive positioning, is the right one, and his journey across Silicon Valley, Toronto, Las Vegas, and New York to test it gave the talk authority. His framing of GEO and AEO as "some of the most important new business conversations the industry has seen in years" closed the conference on the note it deserved.

Janus Boye wrapped with the interactive what-is-ahead session, which is always the best part of a Boye conference because it is genuinely interactive and the room is by then comfortable enough to push back.

A side note: MX is not GEO

Several talks at this conference touched, directly or implicitly, on the question of how to be visible to AI. Nicole Rogers named GEO, AEO, and AIO. Matt McQueeny built his keynote around it. Søren spoke about the shift to fewer clicks but higher intent. Kate Kenyon framed it as the gap between content creation and content infrastructure. My own talk was about the new audience for the web.

These are all variants of the same conversation, and the conversation is the right one. But there is a distinction worth being precise about, because it materially affects what a company should invest in.

GEO, Generative Engine Optimization, is a marketing discipline. It asks one question: how do I get an AI to cite my page? It produces real, measurable outcomes: citation rate dashboards, Share of Model metrics, content tactics that move the needle. The brands practising it are seeing results. It is not a fad.

MX, Machine Experience, is a broader question. It asks: can any machine, crawler, assistant, robot, autonomous system, find any document in a corpus, verify it is genuine, and know whether it is current? GEO tunes one pathway. MX builds coverage across every pathway, because which pathway a given machine will use is unknowable in advance.

The two are not in opposition. As I put it in MX: The Protocols: GEO improvement is an outcome of implementing MX. MX is not a kind of GEO. The relationship is asymmetric. When you do the MX work, semantic HTML, machine-readable metadata, server-side rendering, structured pricing markup, cryptographic provenance through REGINALD, the GEO improvement happens automatically as a by-product at the Citation level. The reverse is not true. A site that has done excellent GEO work without addressing the underlying machine-readability still fails the moment the consuming agent changes, when it becomes a vision-capable agent reasoning from screenshots, an audio-LLM agent like Alexa or Siri, an autonomous commerce agent using ACP or UCP, or simply an impatient crawler with a tight timeout that aborts before the page has finished assembling.

There is also a regulatory dimension that GEO does not engage with and MX is built to support. The European Accessibility Act began enforcement in June 2025; live court cases are underway. The EU AI Act sits alongside it with its own documentation, logging, transparency, and post-market monitoring obligations. Emerging digital-records legislation is converging on a similar shape. GEO produces no evidence relevant to any of this. A note worth making precisely, because it matters to compliance teams and auditors: MX and Reginald do not grant compliance with any of these regulations, that is a legal duty of the organisation. What they do is make the documentation the organisation must produce structured, machine-readable, tamper-evident, and verifiable on request. The same engineering work, semantic HTML, programmatic form labels, DOM-reflected state, signed documents, supports all three regulatory regimes simultaneously, which is the practical reason to do the work once rather than three times.

For anyone at the conference whose team is currently weighing investment in GEO or AEO services, the honest framing is this: do GEO if you have the marketing budget and the editorial capacity to refresh content every couple of weeks. But do not assume it is a substitute for the underlying infrastructure work, and do not assume it is contributing to regulatory compliance. It is not. MX is the framework that catalogues that underlying work, grades it on a 0 to 5 Readiness scale, and extends it past Citation into the comparison, transaction, and provenance layers GEO does not reach.

I have written this up at more length in MX: The Handbook (out April 2026) and MX: The Protocols (July 2026). For anyone who wants the short version, the framing in one line: GEO tunes a single pathway; MX builds coverage across every pathway.

Final thanks

To Janus for the curation and the discipline of the format; neither has slipped over the years we have been doing this. To Matt Garrepy for the steady hand on both days. To every speaker for the work that went into the preparation, which is always more than the audience sees. To Antonia, Markus, and Matt McQueeny for judging CMS Idol with the seriousness it deserves and the lightness it requires. To Dr. Engel and the Museum für Kommunikation Frankfurt for the venue. And to everyone who came up afterwards to argue, agree, or simply continue the conversation; that is what these events are actually for.

See you at the next one.