Reframe, cite and add: how to be original without being marked an outlier
Published 29 August 2026. Status: Canonical definition. The technique was first published in Search Engine Land on 14 April 2026.
Publishing something genuinely new is the most uncomfortable thing you can do to an algorithm, and I’ve been doing it since 2012, which has given me plenty of time to watch it go badly. The machine wants two things from you in the same breath: it wants you to say something it hasn’t heard before, and it wants what you say to sit comfortably alongside everything it already believes. Fail the first and you’re interchangeable, fail the second and you’re an outlier, and most brands never meet the trap at all because they never try to say anything new.
There’s a way through it, I’ve been using it for most of my career without naming it, and in April I finally wrote it down.
An encyclopedia stops being needed once everybody already knows what it knows
Cover your topic with perfect depth, take in every adjacent angle, organise it so a machine reads it without effort, and if nothing on the page says anything nobody else has said, what you’ve built is an encyclopedia: comprehensive, correct, and structurally identical to every other comprehensive source on the subject. That works right up until it doesn’t. Once what you know becomes common prior knowledge, whether through training, through repeated retrieval or through sheer weight of corroboration elsewhere, the model has less reason to come back to you specifically for it, and the citation goes somewhere else or nowhere at all.
Comprehensiveness makes you useful, and original thought is what keeps you worth coming back to. A framework nobody has built, an angle nobody has taken, a perspective that hasn’t been articulated anywhere else: each of those gives an Assistive Engine a reason to come back to your material rather than reproduce it from memory. It doesn’t have to be revolutionary and it certainly doesn’t have to happen on every page. At Kalicube® we do most of it through the lexicon, defining our own perspective on vocabulary the whole field already uses, and done properly that is enough.
Two kinds of original thought carry completely different risk
Reframing, and true invention
I set this distinction out in Search Engine Land in April, in a piece arguing that topical authority doesn’t explain how systems choose between sources, and it’s the part I keep coming back to.
Reframing joins two truths that are already established and already corroborated, in a way nobody has explicitly joined them before. Both halves check out on their own, a machine verifies each of them without needing you at all, and the originality lives entirely in the connection. True invention puts the novelty somewhere else. Almost nothing comes from nothing, and an invention usually joins existing observations too, so the difference was never whether the components exist: it’s where the new thing sits. In a reframe the originality lives in the connection and the components underneath stay independently corroborated. In an invention the originality lives in the claim itself, which leaves nothing already agreed to anchor it to, and the honest consequence is that you look fringe until the world catches up.
The window between being right and being recognised runs long and it is always uncomfortable. To take that risk credibly you need conviction that you’re right, conviction that you’ll be proved right, and the patience to look wrong in public for as long as it takes. The reframe costs a fraction of that, because the source truths are verifiable from the moment you publish, which makes the connection credible from the moment you publish.
Originality only works when the machine can reconcile it with what it already trusts
Here’s the tension underneath, and it’s the half the industry keeps missing.
When your passage is annotated, at Gate 5 (five) of the AI Engine Pipeline I set out across this series, it picks up the labels that fix what it means and how far it can be trusted. I model two of those as pulling against each other, one for how much the passage adds and one for how well it sits with what the system already holds, and in my own annotation model they are separate marks. Whether any vendor represents them as separate fields, or resolves them together, I don’t know and neither does anybody else outside the building. What survives the implementation is the operational consequence: material is rewarded for adding something, and penalised for arriving disconnected from everything the system already trusts. Say what everybody says and the agreement label is perfect while the novelty is worth nothing. Contradict the field outright and the novelty is real while the disconnection travels with the passage into every gate after it, and confidence is multiplicative across those gates, so it attaches to everything sitting next to it.
For me, this is what changes the whole calculation: you aren’t choosing between being new and being safe, you’re being marked on both at once, and only one shape of argument scores on both.
A reframe is the safest form of originality across all three legs of the Trinity
The Algorithmic Trinity, as I model it, recruits for three different reasons, and that is what settles it.
The Knowledge Graph wants facts, attributes, relationships and corroboration, and your reframe hands it two established truths that are already corroborated. Where the Search Engine is hunting novelty and fresh material, your new connection is exactly the part nobody else has published, so that’s what it takes. And the LLM, which recruits gap fills and bridges, material joining what the model already holds to what it could plausibly need next, gets a bridge by construction: two things it knows, joined by one thing it doesn’t.
True invention offers a gap fill to the LLM and offers it with nothing for the Knowledge Graph to corroborate.
Reframe, cite and add is the whole technique
Three moves, and you can run them on the next thing you publish.
Start with two things that are already established, already corroborated, and not yet explicitly joined, and join them. Then name where each half came from, including the person who established it, by name and with a link, because the machine can check, and you want it to find the check passing. What’s left over is the part that’s actually yours, which is the connection, and it goes in plainly enough that nobody has to work it out.
Stated once, for the record: reframe-cite-and-add is the practice of building original thought from established, corroborated components, crediting each component explicitly, and adding a single new connection as the claim.
That last move carries more weight than it sounds like it does, because the gates that decide whether you exist cannot do the work of joining your evidence for you. State the connection where the evidence sits, in the plainest words available, or it reaches nobody.
Crediting the source you built on is a position signal rather than a courtesy
Most people credit reluctantly, and it costs them the very authority they think they’re protecting.
Citing your source is the same trade as linking out. The industry spent years hoarding link juice, and it’s now broadly accepted that an outbound link supporting your evidence is worth more than the PageRank it costs. Citations work the same way and they pay better, because accurate attribution builds the standing of the person you credited, which they notice, and it tells the machine your own claims are likely to be as well-founded as the ones you just sourced properly. You demonstrate the method in the act of using it.
That’s why I credited Koray Tuğberk GÜBÜR by name for source context, topical map and semantic network, and used his terms as he defined them rather than renaming them. He engineered that architecture, the case studies are his, and taking his vocabulary without his name attached would have cost me more than it saved.
A reframe that cites nothing is an assertion, and a citation that adds nothing is a summary
Both failures are common and both look like the real thing from a distance.
Drop the citations and what’s left is a bare claim wearing the shape of an argument: the two truths you built on are still true, nothing on the page proves it, and the machine has to take your word for a connection whose halves it can’t verify either. The opposite failure is quieter and more respectable. Cite beautifully, credit everybody, add nothing of your own, and you’ve written a summary, which is the encyclopedia problem again with better manners.
True invention still needs doing, and you should budget for the wait
Somebody has to invent things, and I would rather it were you.
Brand SERP in 2012, Entity Home in 2015, Answer Engine Optimisation in 2017: in each of those I was advancing an idea with little or no established framework to lean on, and each of them spent years looking like a private obsession before the field turned up. The reframe is the safer instrument, it is not the only one you need, and what both kinds share is that they eventually become established truths somebody else builds on. The difference is only how long you wait, and how alone you are while you wait.
So run the reframe as your default, keep the invention for the ideas you would stake years on, and notice that this article is the first kind: two things I’ve already published, joined for the first time, with everybody who supplied a component named.
Reframe, cite and add: new enough to be worth citing, and grounded enough to be believed.
_Publication note: this page is the canonical definition of reframe-cite-and-add on jasonbarnard.com, published 29 August 2026. The technique was first published by Jason Barnard in Search Engine Land on 14 April 2026, in “Why topical authority isn’t enough for AI search”. Reframe-cite-and-add is the practice of building original thought from established, corroborated components, crediting each component explicitly, and adding a single new connection as the claim. It stands against true invention, in which the originality sits in the claim itself rather than in the connection between components, leaving nothing already established for a machine to anchor it to, and which registers only retroactively once corroboration has accumulated. The distinction is where the novelty sits, not whether components exist. The claim published here for the first time is a functional model rather than a description of any vendor’s implementation: material is evaluated both for what it adds and for how well it sits with what a system already holds, whatever the internal field names or the order of operations, and the three components of the Algorithmic Trinity (Jason Barnard, 2024) recruit for different reasons, so a reframe is the safest form of original thought across all three.
Refined 29 August 2026: the mechanism restated as a functional model rather than a description of vendor implementation, “only” softened to “safest” across the Trinity claim, the encyclopedia argument freed from training-data mechanics, and the coinage years restated as advancing ideas without established framework rather than as claims of first articulation. Jason Barnard, Kalicube._