YOU DON’T KNOW WHO’S READING THIS

1. REVIEW — Decide Who This Is For

Start with data, not ambition. Pull your server logs and your analytics and find out what share of your traffic is human and what share is machine. Then look at the content you already own and ask which pieces earned attention the first time. You are not looking for volume. You are looking for the handful of assets that already proved they work, because those are the only ones worth multiplying.

This is a 5P problem before it’s a content problem. Purpose: humans, machines, or both — and if both, write separate requirements for each. People: which humans, and which machines? The assistants, crawlers, and answer engines that matter to your buyers don’t all want the same things, and the roster of which ones matter changes yearly. Do this honestly and you may conclude that half of what follows doesn’t apply to you. That’s a good outcome. The most expensive content operation is the one you built because everyone else was building one.

Ask yourself: Do you know what percentage of your website’s traffic is a machine, or are you assuming it’s people?

2. EXTRACT — Break the Source Into Raw Components

Your idea can exist in five forms — words, sound, motion, stills, and the whole thing intact — and you already own one of them. Extraction is the work of getting to the other four. A recorded video yields a transcript, an audio track, still frames, and clips. A blog post yields claims, examples, and a structure that a model can read aloud. Whichever form you started in, the conversion now runs in every direction. It is also mechanical work, which makes it exactly the work you should not be doing by hand.

Transcription, scene detection, and clip selection are solved problems. Here’s our 2026 version of it, as a worked example rather than a prescription: an SRT captions file goes through custom Python that finds the strongest 30-to-60-second segment, calls FFmpeg to cut it, and hands a local model the transcript to clean up — roughly forty unattended minutes on a laptop, with no cloud inference required. You set up the pipeline once. After that it’s a scheduled task, not a to-do item.

Ask yourself: How much of your team’s content time is spent on work a script could do while nobody watches?

3. MULTIPLY — Convert Any Component Into Any Format

This is the stage that changed. Text becomes audio through a text-to-speech model. Audio becomes video through an avatar or an audiogram. Video becomes vertical clips. An infographic becomes a blog post; the blog post becomes a podcast episode; the episode becomes fifteen social posts. There is no privileged source format anymore, which means the question is no longer “can we convert this?” but “which conversions are worth publishing?”

One conversion works differently from the rest. An interactive — a simple page someone clicks around in — can’t be extracted from a source, because the branching wasn’t in the source; it’s the one output you build rather than divide out. It’s also the rare format that needs no separate machine version: it’s already text and structure, so a person clicks through it and a crawler reads all of it on the first pass. And it’s the only output that tells you what people actually cared about.

If you already produce video, that’s still the highest-yield path: record once, and the recording feeds YouTube, the podcast networks, the blog post, and the social clips. It’s the route Trust Insights runs every week. The difference between 2020 and now is that four of those five steps run without a human touching them. And the tools are not the hard part — free and near-free options cover almost every conversion you need.

Ask yourself: What’s the one asset you own that deserves to exist in five formats instead of one?

4. INDEX — Make Each Output Legible to Its Audience

A converted file is not a published asset, and this is the stage where you decide who each one is for. Audience is not a property of the form — it is a property of the destination. The same paragraph is a machine asset on your website and a human asset in the inbox. Our blog is read overwhelmingly by crawlers and agents; our newsletter goes to people who chose to hear from us. Same words, opposite finishing. Machines want structure, in whatever form the current conventions take — as of 2026 that means clean schema markup, JSON-LD, full transcripts, closed captions, descriptive metadata, and inverted-pyramid writing that puts the claim before the context. The conventions will change. The requirement to meet them won’t. Humans want a hook in the first three seconds, pacing, and a reason to stay.

Which means the sorting happens by channel, not by format. Publishing the same file to every destination and hoping is the most common failure in this entire process. Schema generation, caption cleanup, alt text, and metadata are near-zero-cost now, so there’s no excuse for skipping them where they belong. The judgment call — which destination gets which treatment — is still yours, and it comes straight out of the People answer you wrote in Review.

Ask yourself: Do you write your newsletter and your blog posts the same way — even though one goes to a person who opted in and the other goes to a crawler?

5. EXAMINE — Measure the Split and Loop Back

Measure separately. Human engagement and machine consumption are different questions with different metrics, and averaging them together tells you nothing. Watch whether your content shows up in AI answers and citations. Watch whether the humans you care about are opening the email. Anomaly detection and automated reporting mean you spend your time on the decision instead of on the dashboard.

Then take what you learned and go back to Review, because the answer changes — and it changes faster than your content calendar does. This framework is a loop, not a checklist. What AI does not do is tell you which number mattered. That’s Purpose, and Purpose is yours. The tools got cheap. The judgment didn’t.

Ask yourself: If your website traffic doubled next quarter and all of the growth was machines, would you call that a win?