Kindling: the writing tool that refuses to write your post

Kindling does not write your post. It shows you two opposite takes on a topic from your field, and both are labeled "this is not your voice." You react in your own language, typed or dictated. The post is built from that reaction, in your style, and then carried into English. You edit it, and you publish it yourself.

That is the whole product in one paragraph. The rest of this piece covers why we built it that way, what it refuses to do, and the numbers so far.

Who it is for

A compliance officer, an analyst, an engineer, a lawyer, a broker. Mid-career, good at the work, often working outside their native language. They know a LinkedIn presence affects their career. They still don't post.

In our experience the reason is almost never laziness. The blank box is intimidating, their own opinion feels too obvious to say, and their English doesn't feel good enough.

We think text generators don't solve that problem. They cover it up. A post a model wrote sounds like a thousand other posts a model wrote. The author's voice disappears, and a personal brand is built on exactly that voice.

What readers actually flag

Two outside signals. Neither is our research.

A LinkedIn post from 16 September 2026 (264 reactions, 101 comments) argued that B2B buyers have learned to spot AI-written content. Here is the sharpest reply in our own paraphrase: what readers pick up on isn't AI. It is the absence of a specific position. A post in generic phrasing gets flagged as machine-written even when a person wrote it.

A 2026 roundup of AI-in-social-media statistics from Memeburn (sample not stated in the roundup) separates "AI wrote it" from "AI helped." Fully AI-generated posts underperform human-written ones by 2% on LinkedIn, 3% on X and 6% on Instagram. Posts where AI only assisted outperform human-written ones by 5% on LinkedIn.

Our reading: the gaps are small, and a roundup is not a controlled study. But it points the same way as that comment. The penalty isn't for touching AI. It is for publishing a post with nobody's opinion in it. That is the problem Kindling is built around: its first job is to get the opinion out of the person, before any text exists.

How it works

The method is Socratic: provoke first, then help phrase. Think of it as a small PR agency in your pocket.

1. feed      topics from your profession,
             not a general stream about AI
2. provoke   two deliberately polar drafts on the chosen topic,
             labeled "this is not your voice"
3. react     you answer in your own words, in your own language;
             dictation works
4. build     a post in your style, in that language,
             then carried into English with the intonation kept
5. publish   you edit, you post; there is no autopublishing

Step 2 is the one that needs explaining. The two drafts are not options to pick from. They exist to get a reaction: "no, that's not how it works," or "yes, but I'd say it harder." That reaction is the real opinion. It's raw, but it's yours. Everything after step 3 shapes what you said. It never adds a view you didn't express.

Rules the product won't break for a metric

- no score: never "voice match 88%"; writing is not an exam
- never invent the author's opinion or experience
- never read the author's likes or reach;
  one feedback button: worked / didn't
- no change to the style profile or mode happens silently
- no autopublishing

Each rule costs something, and we pay it on purpose. A score would be easy to show and would look like progress. We think a number on screen pushes people to write for the meter instead of saying what they mean. Autopublishing would save a click. It would also mean a post could go out that the author never read in its final form. Reading likes and reach would give the product more signal. Instead the author decides whether a post worked, with one button.

It learns from your edits

The first post is only the start. Kindling learns from what you change. If you keep softening categorical statements or cutting bureaucratic phrasing, it notices the pattern, offers to remember it, and shows you what changed in your style profile. Nothing is rewritten silently.

As the author finds their voice, the provocations get gentler. Experienced authors get three different angles on a topic instead of two extremes.

Skipping costs nothing

The gesture people repeat most is skipping a prompt they don't care about. We made sure that gesture is free.

visible feed    a window of 7 prompts
pool behind it  24 prompts
skip a card     the pool rotates; no new material is bought
budget spent    the feed says so instead of quietly shrinking

The user never feels this machinery. They just see a feed that doesn't shrink when they skip cards.

The numbers so far

first line of code     29 May 2026
code changes since     570+ in four months
input languages        13, including Arabic, Farsi and Hindi
output                 English

The Kindling case page reports one result: the first post built this way reached 1,708 impressions. That is one author's result, and the page presents it as one author's result. It is not a promise. We don't promise viral posts. We promise that every text you publish is yours.

What we traded away

This is our judgment, not a measurement. Kindling is slower than a generator by design, because you have to react before there is anything to edit. Some people want the text handed to them, and this is not that tool. No score means the interface gives less of a feeling of progress. No autopublishing adds friction at the end. We chose all three because each one protects the same thing: the post on your profile says what you think.

FAQ

Does Kindling write the post for me? No. It writes two deliberately opposite drafts that are labeled as not your voice. Their only job is to get a reaction. The post is built from what you say back, and you edit the final text.

Do I have to think in English? No. Kindling takes your thoughts in 13 languages, including Arabic, Farsi and Hindi. It builds the post in your language first, then carries it into English with your intonation kept.

Will it post to LinkedIn for me? No. There is no autopublishing, and that won't change. You publish every post yourself.

Why is there no voice-match score? We think a percentage turns writing into an exam, and people start performing for the number. Kindling doesn't show one. The only feedback it asks for is one button: worked or didn't.

How is this different from asking a chatbot for a post? A prompt starts from text. Kindling starts from your reaction to two opposing takes, so the opinion exists before the wording does. It also learns from your edits and shows you what it learned. In our view, that is the part a one-off prompt can't do.

Try it

Kindling is live at kindling.iloblique.com. The full case, with the product's invariants and the first result, is at iloblique.com/kindling.

If you have something worth building, we'd like to hear about it.

Sources

• Kindling case page: iloblique.com/kindling
• Memeburn, roundup of AI-in-social-media statistics, 2026 (sample not stated)
• LinkedIn post on B2B buyers and AI-written content, 16 September 2026, with its comment thread; paraphrased, author not named