Most crypto content programmes are calendars. Twelve pieces a month, published on schedule, read by almost nobody, forgotten within a week.
The ones that work publish fewer things that keep earning attention for years.
The metric that tells you which one you have: what share of your traffic comes from content published more than six months ago? If it’s small, you’re filling a calendar. If it’s growing, you have an engine.
Fewer, better, and built to last
A piece that answers a real question completely will get traffic, citations and links for years. A piece commenting on this week’s news has a shelf life measured in hours.
The arithmetic is unforgiving. Three genuinely useful pieces a month, each still working two years later, accumulate into an asset. Twelve forgettable ones a month accumulate into a maintenance burden.
This is unpopular advice because volume feels like progress and is easy to report. Depth is harder to schedule and harder to defend in a status meeting.
What earns attention in this market
Complete answers to real questions. Not overviews. The piece that finally explains the thing properly, with the edge cases and the parts that usually get skipped.
Post-incident analysis. When something breaks in the market, a clear technical breakdown is the most-shared content in crypto. If you have relevant expertise, this is the highest-return format available and the window is short.
Original data. Anything derived from on-chain analysis nobody else has run. Dune makes this accessible and remarkably few companies do it.
Comparisons that take a position. Not a balanced table with no conclusion. An actual recommendation with reasoning, including who should choose the other option.
Technical writing adjacent to documentation. How to actually do the thing, with working code. Developers share it and it ranks forever.
What doesn’t
News commentary. By the time you publish, the conversation moved.
Opinion pieces containing no specific claim. If a reader can’t disagree with it, they also can’t learn from it.
Keyword-first content written by someone who researched the topic that morning. In this market readers check, and being wrong in public about something technical is expensive.
Anything generated wholesale. It’s characteristically hedged and generic — the exact opposite of what earns citations from both readers and AI systems, which weight specificity and independent corroboration.
Who writes it
The model that works: the person who knows the subject drafts, an editor shapes it, a subject expert reviews before publishing.
Writers researching an unfamiliar technical topic produce content that reads plausibly and gets details wrong. That’s worse than publishing nothing, because a technically literate reader spots it immediately and revises their view of everything else on your site.
Getting engineers to write is a management problem, not a talent problem. Most will if the draft is a conversation transcript rather than a blank page, and if someone else handles the editing.
Where AI fits
As a tool for people who know the subject: genuinely useful. Structuring an outline, drafting a section you’ll rewrite, checking whether you’ve covered the obvious objections, turning a transcript into prose.
As the finished product: no. Generated content hedges, avoids specifics, and reads like every other generated page targeting the same query. It also fails the thing that matters most for AI citation, which is containing specific facts and defensible positions that a model can attribute to you.
The irony is worth noting. Content written to be cited by AI systems has to be more human, not less — more specific, more opinionated, more anchored to things only someone doing the work would know.
Structure for citation
Since a growing share of your readers are systems rather than people:
- Answer in the first sentence under every heading.
- Specific numbers and dates, in absolute form.
- Named entities — chains, tools, standards, companies.
- Lists of three or more items rather than comma-laden sentences.
- Tables for anything comparative, which get reproduced with unusually high fidelity.
- Definitions written as definitions.
This is also just clearer writing, which is why it works for both audiences.
Measuring it
- Traffic from content older than six months, and whether that share is growing.
- Citations — by AI assistants when you run your priority questions, and by other publications linking to you.
- Branded search growth, which tends to follow content that circulates.
- Conversions from content-sourced sessions, tracked properly rather than assumed.
Publishing volume and social engagement are activity metrics. They tell you the programme is running, not that it’s working.
What drives the work
- How much internal expertise you can get onto the page, which is the largest variable.
- Technical depth of the subject, since deeper topics take longer and are worth more.
- Number of languages, since translation isn’t the same as writing for a market.
- Review requirements, particularly where content touches claims your legal team has to clear.
- Existing content inventory, since rewriting what you have is frequently higher-return than publishing something new.
The recommendation
Cut your publishing volume. Put the saved effort into making each piece the best thing on the internet about that specific question.
Then go back and rewrite your existing pages to the same standard, because the ones that already rank are the ones with the most to gain.
Common questions
How much content should a crypto company publish?
Less than most content calendars suggest, and better. A programme producing three genuinely useful pieces a month that keep earning traffic and citations outperforms one producing twelve forgettable ones. The measure worth tracking is how much of your traffic comes from content published more than six months ago, because that is the number that tells you whether anything is compounding.
Who should write technical crypto content?
People who understand the subject, edited by people who can write. The reverse — writers researching a topic they do not understand — produces content that reads plausibly and gets details wrong, which is worse than publishing nothing in a market where readers check. The practical model is engineer or operator drafts, editor shapes, subject expert reviews before publishing.
Does AI-written content work for crypto companies?
As a drafting and structuring aid, yes. As the finished product, no. Generated content is characteristically hedged, generic and free of the specific numbers, named entities and opinions that make a page worth citing. It also increasingly competes against every other generated page saying the same thing. Use it to accelerate people who know the subject, not to replace them.
What content formats work best in crypto?
Explainers that answer a real question completely, post-incident analysis, original data work, comparison pieces that take a position, and documentation-adjacent technical writing. What underperforms is news commentary, which has a shelf life of hours, and opinion pieces that do not contain any specific claim a reader could disagree with.
How do you measure a content programme?
Traffic from pieces older than six months, citations by AI assistants and other publications, branded search growth, and conversions attributed to content-sourced sessions. Publishing volume and social engagement are activity metrics. If the programme is working, an increasing share of your traffic should come from things you published a long time ago.
