Answers
Fixing an unsourced claim: the RankAI method
How do you fix an article when a claim has no source? Check whether the fact is in the authoritative record. If it is only partly documented, turn it into an explicit limit. If it is not documented at all, cut it.
How do you fix an article when a claim has no source? The short answer: check whether the fact is in the authoritative record. If it is not there, turn it into an explicit limit when it is partly documentable. If it is not documentable at all, cut it. The RankAI method below is a working proposal, not a universal standard. Take a negative example: writing “RankAI measures 豆包” is a mistake to correct, not a description of the product. The same goes for 元宝, Kimi or 千问: a page in simplified Chinese does not mean RankAI measures those engines. The correction starts from the question, not from the example.
The method in four steps
The RankAI procedure is meant for people who write or review content and come across a claim that does not hold up. It is not a universal rule: it is a way of working that we make explicit.
Step 1 — Isolate the claim
Find the single sentence that lacks support. Do not rewrite the whole paragraph: isolate the fact and ask yourself whether it is already in the record of things we can state. If it is not, it does not go into the text.
Step 2 — Check the record
Check whether the fact is among those documented. If it is not, move to step 3. If it is, rewrite it using the wording closest to the record, without adding details.
Step 3 — Choose between limit and cut
Here you need a binary criterion: if the fact is partly documentable, turn it into an explicit limit; if it is not documentable at all, cut the sentence. Example: “RankAI measures 豆包” is not partly documentable, so it gets cut. A visibility measurement that is missing, by contrast, is a documentable limit: the product shows it as N/A, never as zero.
Step 4 — Reread in context
After the correction, reread the passage to make sure no implicit references to the removed fact are left. If the text was about questions on ChatGPT and Google, check that the correction does not shift the topic to engines we do not measure.
The delicate point is telling method apart from product fact. The RankAI method — isolate, check, choose, reread — is a working proposal, not a law. The product fact, on the other hand, is a constraint: if a measurement is missing, we show it as N/A, never as zero. Keeping the two levels separate avoids two opposite mistakes: turning an editorial choice into a universal rule, or presenting a product limit as if it were a style guideline. In the context of questions on ChatGPT and Google, this separation helps: an API answer is not a person's screen, and Gemini is not Google AI Overviews. These are limits of observation, not sections to add if the question does not call for them. Fixing an unsourced claim comes down to exactly this: working out whether we are talking about what the product does or about how we describe it.
Four-step mini-checklist
- Isolate the unsupported sentence, without rewriting the whole paragraph.
- Check whether the fact is in the authoritative record.
- Choose: partly documentable → limit; not documentable → cut.
- Reread the context to remove implicit references.
Verifiable limit: this method does not establish what RankAI measures; it only establishes how to correct a claim that has no source. For the boundaries of the product, the record still applies: a page in simplified Chinese does not imply that 豆包, 元宝, Kimi or 千问 are measured. Before launch, checkout can stay off: the invitation is to create an account or open the relevant page, not an immediate trial. For more, see the in-depth page.