Your product page
- Weight limit
- 3.5–15 kg
- Newborn insert
- Included
- Last updated
- This season
Measure first. Everything else is opinion.
Four stages, on the three windows below. The first stage produces a number, and every claim after it is answerable against that number.
Month 0
200 observations
Measure
40 questions across 5 engines. Your baseline, and the instrument every later month is read against.
Month 1–2
30–60 days to move
Fix
Access, entity and liftable passages. On-site work depends on nobody else, so it lands first.
Month 3–6
3–6 months
Corroborate
Getting the facts right where the model reads. Slower, because it needs other people to publish or correct something.
Windows we scope engagements against, not measured averages — Searchd has run no client panels.
Stage 01
Build the panel
Forty questions your buyers actually ask, fixed for twelve months, run across five engines. Every run records whether you were named, the exact sentence, who was named instead, and every source cited.
40%30%20%10%
40%Category — “best X for Y”
30%Attribute — your differentiator
20%Comparison — you vs a rival
10%Brand — about you directly
Brand questions are the smallest slice on purpose. Being named when someone already typed your name is not the commercial question.
Stage 02
Diagnose the gap
We diagnose five causes of not being named. The fixes are unrelated, so knowing which one applies is most of the job.
| Order | Cause we check | Time to rule out |
|---|---|---|
| 01 | Crawlers blockedA bot rule returns 403, or the page needs JavaScript to show its text. | 5 min |
| 02 | Nothing quotableReadable, but no paragraph survives being pulled out of the page. | 15 min |
| 03 | Entity unresolvedYour name exists in several forms. None of them wins. | 1 hour |
| 04 | Nobody corroboratesOnly your own site states your specification. | half a day |
| 05 | Competitor owns corpusYou are absent from the reading list, not from the market. | days |
Separately from whether you are named: what is being said. A wrong SPF value, a wrong weight limit, a wrong age range — stated with the same confidence as a right one. For a regulated product that lands on your compliance owner’s desk.
Stage 03
Fix what you own
Fastest movement, because it depends on nobody else.
Cannot be lifted
“This is why it matters so much here, and it is worth understanding the reasoning before you decide.”
No subject. No number. Pulled out of the page, it names nobody. Sample copy, fictional brand — the shape is the point, not the words.
Can be lifted
“Haruwell Daily Fluid is SPF 50+ PA++++, uses chemical filters only, and was tested at the full 2 mg/cm² application rate.”
Names the product, carries three checkable numbers, states the test condition. Sample copy, fictional brand — the values are invented.
- Unblock AI crawlers, confirm from outside your network
- Every page complete in raw HTML
- One entity string everywhere, resolved in schema
- Specification published as quotable fact
Stage 04
Corroborate off-site
Slower, because it depends on other people publishing or correcting something.
Correct the record
Brand registry where you own the brand, documented correction requests citing your own specification page where a distributor controls the listing. Re-checked monthly, because a corrected listing can revert.
Get into the reading list
The comparison pages, round-ups and communities your category is actually answered from. A claim only you make rests on your word. A claim independent sources also make does not.
Distributor’s marketplace listing cited
- Weight limit
- 3.5–15 lb
- Newborn insert
- Sold separately
- Last updated
- Two versions ago
What the assistant then states“It holds 3.5 to 15 pounds, and the newborn insert is a separate purchase.”
Method questions
How do you choose the 40 questions?
From how buyers actually phrase things — Search Console query strings, your own reviews and support tickets, category forums. Not a keyword tool: assistants are asked in sentences. The panel is then fixed for a year so the numbers stay comparable.
Answers vary every time you ask. How is this a measurement?
They do, so we control what can be controlled — same day of month, clean sessions, no personalisation, region pinned — and accept the rest as real variance.
Named rate is reported per engine, so each figure rests on 40 questions, and at that sample size a few points is sampling error. Every engagement will therefore open with a stability run: the same panel, five consecutive days, nothing changed. That spread becomes your noise floor, it goes on every report, and any move inside it is labelled noise rather than progress.
What makes a passage quotable?
It answers one question completely in its first two to four sentences, names its subject instead of saying “it”, and carries a number with a source. Copy written to be read top to bottom fails the first test alone — it is written to be read top to bottom, and models read in chunks.
Why does off-site work matter more here than in SEO?
A model is asserting a fact, not returning a link. A specification only your own site states is a claim it may decline to repeat, because the only source for it is the party who benefits from it. Independent agreement is what removes that problem.
Find out what AI says about your brand.
Send us your brand and category. We run 20 buyer questions across five engines by hand and send you the transcript within 48 hours.
- What arrives
- 20 questions, 5 engines, every answer quoted in full
- Who sends it
- Snow Lee, by hand, from Seoul
- Your email
- Used to send the report and reply about it. Privacy.
| Buyer question | Engine | Named |
|---|---|---|
| best Korean sunscreen for oily skin | ChatGPT | not named |
| K-beauty sunscreen that does not pill | Claude | named |
| Korean skincare brands sold at Ulta | AI Overviews | not named |
| …17 more questions, with the verbatim answer and every source cited | ||