Two Mirrors: See the Link Clearly, Then Reflect the Method Back on Yourself

When you like a website, don't rush to copy it

A mature website easily creates an illusion: just move over its colors, cards and articles, and your own site will instantly look mature too.

Beyond the pages, there's actually the content scale built from long-term updates, the pitfalls the author personally hit, accumulated search presence, asset licensing and reader trust. Take only the appearance and you get an empty shell; take the body text and images too, and you pass off someone else's experience, evidence and risk as your own.

Now I force myself to look through two mirrors in a row. The first only confirms what the other side actually did; the second asks what this mechanism, put back into my goal, should become — adopt, adapt, wait, or reject outright.

The order can't be reversed. Without first seeing the source clearly, so-called learning easily slides into imitation.

The first mirror only does forensics, not learning

When reading a link, I first break the page into verifiable objects.

Page identity

Record the exact URL, title, author, publish time, update time, section and license. A title in search results may be truncated, and the summary may not be the author's own sentence, so you can't substitute a search page for the original.

Content type

Distinguish whether this is a tutorial, an opinion, news, promotion, personal experience or a resource roundup. Different types need different evidence.

A tutorial needs its commands and versions verified; news needs the event timing verified; a promotional piece needs its conflict of interest identified; personal experience can't be rewritten directly as your own.

Facts versus expression

"Feature X exists in version Y" may be a fact. "This is the best option" is usually a judgment. "I've used it for six months" is the author's experience.

The three can't be moved together into a new article just because they appear in the same paragraph.

License and assets

A body-text license doesn't automatically cover the cover, third-party screenshots, embedded video, trademarks and comments. Even if a page declares some open license, check the limits on commercial use, attribution, share-alike and third-party assets.

Appearing all over the web only proves it spread widely — not that it's copyright-free.

The second mirror asks: what's this worth for my goal

Only after seeing the original clearly do I return to my own system. Each method has to pass five questions:

1. Is the underlying problem it solves the same as the problem I have now.
2. Do I have the team, budget, accounts and traffic conditions it depends on.
3. Which mechanism can be verified on its own, without copying the whole set.
4. Will it conflict with my security, brand or business boundaries.
5. How can a minimal experiment give feedback within limited time.

The second mirror's output isn't a summary — it's one of four decisions.

Adopt: the mechanism matches the goal and can go straight into the existing flow.

Adapt: the underlying idea is useful, but it needs different evidence, structure or applicable scope.

Wait: the value may exist, but the conditions to verify it are currently missing.

Reject: even if effective, it doesn't fit the goal or the risk boundary.

"Bookmark" isn't a fifth decision. A bookmark with no next step is just deferred judgment.

A full example, using this blog's redesign

The reference blog's title-based cover solved a real problem: when a user scans cards on the homepage, they quickly know what an article is about.

The first mirror told me it uses landscape images, with the title taking the main visual, a background related to the topic, forming a continuous content density on the homepage.

The second mirror then asked me to keep judging:

• Do I need to copy its images? No.
• Do I need to copy its fonts and colors? No.
• Is a title as the cover's core right for my readers? Yes.
• What's still missing from my articles? Difficulty, time, verification status and failure notes.
• Is the business goal the same? No — I need to route traffic to Content Factory and workflow services, but I can't let every piece become an ad.

What I adopted in the end was the "title-first cover mechanism," what I redid was the brand, images and information structure, and what I rejected was moving the original images and body text.

That's borrowing the mechanism, not the finished thing.

What a link card should leave behind

To stop AI from kneading everything into one smooth summary, I have each link leave at least five blocks:

source
  original URL, author, date, license

claims
  verifiable facts and where the evidence sits

unknowns
  what the page didn't make clear or what can't be confirmed now

decision
  adopt, adapt, wait or reject, and why

action
  which one concrete change to make to the existing system

If all that's left at the end is "this article is inspiring, worth learning from," the distillation isn't finished.

An adversarial check: assume this content is misleading me

First-principles thinking helps me find the mechanism; the adversarial check is responsible for finding blind spots. I ask in reverse:

• If the title exaggerated, what did the body actually deliver?
• If the screenshot is outdated, does the current interface still exist?
• If the author has promotional income, could the recommendation order be affected?
• If the tutorial only shows the success path, who bears the failure cost?
• If I don't have the author's account, region or device, does the conclusion still hold?
• If you strip the brand and the story, what verifiable value does the method still have?
• If search engines rank several retelling pages first, where is the upstream source?

These questions aren't to negate all external content — they're to stop "someone else pulled it off" from being mistranslated into "I'll surely pull it off by copying."

What AI is best at in this chain

AI is great at extracting page structure, listing claims to verify, comparing several sources and generating research questions.

It's not suited to make these judgments for me:

• Whether the license is enough to support commercial use.
• Whether the author's experience can migrate into my experience.
• Whether the current version actually runs in my environment.
• Whether a given risk is worth taking.
• Whether the final view is Kenton's real judgment.

The model can help me see the material faster, but it can't own the evidence and responsibility for me.

Why "paragraph-by-paragraph rewriting" isn't original reconstruction

If you open the original and swap synonyms from the first paragraph, even if every sentence changes, the article may still keep the original author's unique structure, choice of order and narrative design. It also lets your verification be led along by the original's path.

A safer method is:

1. Extract only the problem it solves and the search intent.
2. Close the original body.
3. Go back to the official docs, the original repo and the current product interface.
4. Re-verify in your own environment.
5. Build the structure from a blank page.
6. Use your own screenshots, failure notes and conclusions.
7. After finishing, only do a fact cross-check — no paragraph-by-paragraph comparison rewrite.

The article you get this way may reach facts similar to the reference, but the evidence chain, structure and expression belong to your own production process.

The two mirrors leave one decision in the end

The first mirror stops me from misreading others; the second stops me from losing myself while learning from them.

For a long-term content system, the most important thing isn't collecting the most links — it's that every external input leaves a clear boundary: what's the original source, what's a reusable fact, what's Kenton's judgment, and what finally changed the system.

Next time you hit a website you want to lift wholesale, bring back just one mechanism to try. Write down what problem it solves, why it suits you, and what the minimal verification is. Until those three are clear, don't copy the finished thing.