The interest graph is no longer enough: why your website matters more than it has in ten years

For ten years the advice has been unambiguous: go where the people are. Build on social media, let the website be a brochure. That advice was right for as long as attention was the scarce good. Now attention is no longer the only thing that decides. What decides is who gets cited when someone asks an AI model a question, and there the channels you have spent ten years building count for almost nothing.

What actually happened to the interest graph

Until around 2016, social media was mostly a social graph: you saw content from people you followed. Your reach was in practice limited by how many had chosen you.

Then the platforms changed engines. TikTok showed that content could be ranked by interest rather than relationship: show a video to people who would probably like it, whether or not they followed the account. Instagram and the others followed, and the result was dramatic. A small account could suddenly reach enormous numbers without owning an audience in advance.

It was a real opportunity, and those who understood it early got a great deal out of it. The consequence was equally real: budgets, people and attention moved from owned surfaces to the platforms. The website became something you updated when someone complained.

I was on the inside of that shift

I was digital lead for the communications department at NATO Allied Air Command when this happened, and I took that bet on behalf of an organisation that is not permitted to buy reach.

In a best-practice presentation given to NATO's communications community in spring 2024, my main message was that Instagram had to be prioritised, not because it was popular, but because changes in how the platform distributed content had made it useful for our purpose. It was an assessment of mechanics, not of fashion.

The results over the period:

• 432 % organic growth over three and a half years, composed of annual growth of 155.5 %, 38.5 % and 20.3 %
• Around 250 million impressions in total, without a single krone of ad spend
• Over 10 % average engagement rate in 2024, higher than the preceding years despite lower publishing volume
• Second of all NATO command channels, behind only NATO HQ's main channel, with the smallest communications team in the structure

The last item on that list is what matters here: engagement went up while activity went down. That is not luck. It is what happens when you stop producing volume and start understanding what the distribution engine actually rewards.

I am not writing this to summarise a former job. I am writing it because I read the last shift correctly and early, and because I think another one is under way now that far fewer have taken on board.

The shift under way now

More and more questions are put not to a search engine but to a model: ChatGPT, Perplexity, Gemini, Copilot, or Google's own AI summaries at the top of the results. The user gets one answer, not ten links.

That moves the decisive question. Before: do we rank high enough to be clicked? Now: are we in the answer at all?

Which makes it interesting where those models source their answers.

The uncomfortable detail

The social channels most organisations have invested in most heavily, such as Instagram, TikTok, Facebook and LinkedIn, are closed surfaces. The content sits behind logins, infinite scroll and technical barriers that in practice make it unavailable to the crawlers AI systems use.

That means something quite concrete. An organisation can have built fifty thousand followers and years of professional content in its channels, and still be entirely absent when someone asks a language model about exactly that field. The content exists. It is simply not readable by the layer that now delivers the answer.

Your followers do not count when the question is asked somewhere your followers are not.

One important exception — and what it really tells us

I want to be precise here, because there is an exception worth understanding rather than skipping.

Reddit is the single most cited source across AI systems, spanning ChatGPT, Gemini, Perplexity and Google's AI summaries. Reddit is unquestionably social media. So the claim that "websites beat social media" is too crude.

But look at what kind of platform it is. Reddit is openly indexable, text-based, organised by topic, and made up of people answering specific questions. That is precisely why it gets cited. And it is the one social platform no organisation controls. You cannot build a Reddit strategy the way you build an Instagram strategy.

The other thing worth noting: even the most cited single source rarely accounts for more than around five per cent of citations on a given platform. The remaining ninety-five per cent spreads across thousands of domains.

That is genuinely good news, and it is the main point of this article. Citation is distributed. Where social media has a winner-takes-all dynamic, the answer layer has a long tail, and a small, well-structured website can be cited on a specialised question where no large player has written anything better.

What makes content citable

The criteria overlap with classic SEO, but they are not the same. Five things matter most:

1. It has to be readable without a login and without JavaScript. Most AI crawlers do not run JavaScript. Content that appears only after the page has loaded does not exist for them.
2. The answer has to come early. Models typically extract the first paragraphs that answer the question. Constructions with a long run-up get skipped.
3. The structure has to be explicit. Headings that are questions, FAQ sections, clear lists, and structured data in JSON-LD that says what the content actually is.
4. There has to be something original. Your own figures, your own method, your own experience. Models have read the generic accounts a thousand times. They cite what is not available elsewhere.
5. It has to be fresh and dated. Where sources conflict, newer material is weighted higher. An undated article from 2023 loses to a dated one from 2026.

What this does not mean

It does not mean social media is over. The interest graph still works, it still creates attention, and for many purposes it is still the fastest route to an audience.

It means the interest graph is no longer sufficient on its own. Social media creates attention. The website decides whether you exist when someone is looking for answers. They solve two different problems, and for ten years many organisations have solved only one of them.

The practical consequence is not to stop doing social media. It is to treat the website as a publishing surface rather than a brochure, and to make sure the professional content that today lives only in a LinkedIn post or an Instagram carousel also exists in an indexable form you own.

Why it is more urgent than it looks

In my NATO presentation from 2024 there was one sentence I repeated every time someone asked how long it would take to build the channels:

It is like starting a business. It takes two to three years.

That applied to social media, and it applies just as much here. Authority in the answer layer is not built by a campaign. It is built by content that has been up, been updated, been linked to and been read over time.

Which means the cost of waiting is not that you lose a few months. The cost is that you lose the head start someone else is building in the meantime, exactly as with the interest graph. Those who waited until it was obvious arrived after it had become difficult.