AI is describing your brand — make sure it gets the facts right
Write the authoritative facts about your brand once, and SearchChamp checks what AI engines say about you against them — surfacing the exact claim an engine got wrong, what’s actually true, and how serious it is. Available on Pro plans and up.
Credit card required · Cancel anytime · 7-day full access
| Page | Clicks | Status |
|---|---|---|
| / | 412 | Improving |
| /blog/ai-search-guide | 286 | Improving |
| /pricing | 204 | Healthy |
| /features/visibility | 171 | Healthy |
- Organic Search46%
- Direct32%
- Referral14%
- AI answers8%
01The new first impression
Buyers now meet you in a paragraph nobody wrote
Somewhere, a buyer is asking an assistant about your category. The answer names you, prices you, and positions you — a summary the model assembled from sources you don't control, read as fact. That paragraph is their first impression of your brand, and nobody is checking it for accuracy.
- The summary is compressed from whatever the engine last read — not from your messaging
- It carries no byline, so it reads as neutral truth
- It never crosses your desk unless you go looking for it
What is Fernwell and how much does it cost?
Fernwell is a team-scheduling platform for agencies and studios, best known for its workload and timeline views. It starts at $62 per seat per month, with a free tier for teams of up to three. The company is remote-first and offers month-to-month billing on every plan.
02Mentions vs correctness
A mention tool hears your name. This checks the claim.
Classic brand tracking answers "who said our name, and in what tone?" — it has nothing to say about whether the claim was true. Brand Integrity Monitoring inverts the question: not how warmly you were mentioned, but whether what an engine told a buyer about you matches the facts.
- A glowing week of coverage can still misquote your price
- Tone is opinion — a founding year, a price, a location are checkable
- Correctness, not coverage, is what a buyer acts on
94%
positive sentiment · all clear
"It starts at $62 per seat per month"
Your fact sheet says
From $24 per seat / month · fernwell.com/pricing
03Your source of truth, drafted for you
The fact sheet writes its own first draft
You don't start from a blank page. SearchChamp reads your brand profile and the pages already crawled from your site, then drafts at least eight facts, capped at thirty — each citing the real page it came from, each carrying a confidence value. The seeder's instruction is explicit: only facts the material supports.
- Every fact carries a
source_urltaken from your real pages — "Do not invent URLs" is in the instructions - Drafted from your business overview, tagline, audience, values and positioning
- Re-seed from my site whenever your site moves on
Brand fact sheet
The facts AI engines are checked against. Confirm auto-seeded facts or edit them to keep this accurate.
Team scheduling software for agencies and studios.
Fernwell Teams — from $24 per seat / month.
2019.
"SOC 2 certified" — not yet true. Never claim it.
04Seven categories
Seven kinds of truth — one is a tripwire
Description, products, founders, location, founding year, claims — the shape of what an engine can get wrong about you. And a seventh that works in reverse: forbidden claims, the things that are not true of you and must never be said. A certification you don't hold. A market you don't serve. A promise you never made.
- Products is the only category with a Price field
forbidden_claiminverts the check — matching it is the failure- A forbidden-claim violation is automatically critical
Description
What you are, in one sentence engines keep reusing.
Products price field
Each product, with the one number engines misquote most.
Founders
Who actually started and runs the company.
Location
Where you are — or the fact that you're everywhere.
Founding year
A single digit off is still a different company history.
Claims
The positive assertions you stand behind, on the record.
Forbidden claims auto-critical
The list that must never be said of you: a certification you don't hold, a market you don't serve, a claim you never made. If an engine asserts one, the finding skips triage — it is critical by definition.
05The machine drafts, a human signs
Auto-seeded is not auto-trusted
Every drafted fact arrives as Seeded — a proposal, not a truth. Your team confirms it, corrects it, or removes it, and only confirmed facts become the reference a check is measured against. The machine saves you the typing; a person signs off on the truth.
- Three statuses per fact: Seeded → Confirmed → Removed
- Only what your team confirms becomes the standard
- Editing requires editor access — sign-off is a real permission, not a formality
Founded in 2019
Remote-first — no physical headquarters
Month-to-month billing on every plan
Offers 24/7 phone support
Editor access is required to edit the fact sheet.
06Every change, on the record
Who changed this fact, and when — answerable
Every edit appends a snapshot to the fact's history: value, price, source URL, status, and who made the change. The trail is append-only, so it can't be rewritten after the fact. And a re-seed that changes nothing writes nothing — no duplicate versions, no noise.
- Value, price, source, status — and the editor's name, every version
- Append-only: history is added to, never rewritten
- A no-change re-seed writes no version at all
Version history
Fernwell Teams — price · every change appends a snapshot
Aug 9 · re-seed ran
Nothing changed — no version written.
v3 · confirmed · J. Elm · Aug 2, 2026
Status Seeded → Confirmed. This value is now the reference.
v2 · edited · M. Chen · Jul 30, 2026
Price corrected $29 → $24 · source fernwell.com/pricing
v1 · seeded · auto-seed · Jul 28, 2026
Drafted from fernwell.com/pricing · confidence 84%
07Four verdicts, not true/false
"Wrong" is a high bar, on purpose
A binary verdict flatters the judge. Every statement an engine makes about you lands in one of four buckets, and the definitions are deliberately conservative — wrong is reserved for a direct contradiction of your sheet, nothing less. Accurate statements aren't discarded either: they're counted, because they're the denominator that makes your score mean something.
wrongonly when a statement directly contradicts a factunverifiablewhen your sheet is simply silent — never guessedaccurateis counted, not thrown away
It matches the fact sheet. Counted toward the score, not discarded.
It conflicts with a fact that may have changed.
It directly contradicts a fact — only then.
The sheet neither confirms nor contradicts it.
Accurate isn't noise — it's the denominator of your score
08The quote has to be real
No verbatim quote, no finding
The judge's instruction is blunt: never invent a quoted span — copy it verbatim from the answer. And it isn't left to goodwill. Any finding whose quote is not an exact substring of the stored answer is rejected outright — not softened, not repaired, dropped. The tool would rather show you nothing than show you a sentence an engine never said.
- Every quoted span is checked as an exact substring of the stored answer
- A paraphrase is dropped, never patched into a match
- Output is strictly schema-validated; a truncated response retries once at a doubled token cap
…Fernwell covers workload planning and timeline views for client teams. Pricing begins at $62 for each seat, billed monthly, and a free tier covers up to three people…
"Pricing begins at $62 for each seat"
Exact substring of the stored answer ✓
"Fernwell charges $62 per seat monthly"
Not a substring — rejected outright ✕
09Critical is reserved
A list that never cries wolf
Critical is spent carefully. It's reserved for a direct contradiction of founding year, price, leadership, or location — or a forbidden-claim violation. Everything else arrives as a warning or as information. A wrong price outranks a loose paraphrase, and when the red pill shows up, it means it.
- Five triggers, spelled out — nothing else goes red
- Warnings for conflicts that may simply have aged
- Info for the merely unverifiable
"Fernwell is SOC 2 certified"
forbidden claim"The mobile app is still in beta"
outdated"Best suited for enterprise teams"
contradicts a claim"Over 4,000 teams use Fernwell"
unverifiable"Integrates with more than 40 tools"
unverifiable"Used by several large agencies"
unverifiable10The finding, end to end
Everything you need to act, in one card
A finding isn't a vibe — it's a case file. The verbatim sentence the engine produced, what your fact sheet says instead, a suggested corrective step, the engine attributed, a severity, a timestamp, and a Dismiss for when it's handled. Dismissal is keyed to that exact statement in that exact answer — the same claim in a new answer returns as a new finding.
- What the AI said — quoted verbatim, engine attributed
- Your fact sheet says — with the source your team confirmed
- Dismiss is per-statement, so nothing stays silently muted forever
Fernwell is team-scheduling software for agencies and studios. It starts at $62 per seat per month. A free tier covers teams of up to three.
Quoted verbatim from this answer
Integrity issues
AI answers that contradict or misstate your fact sheet.
"It starts at $62 per seat per month."
Fernwell Teams — from $24 per seat / month · fernwell.com/pricing
Make the current price unambiguous on the pricing page and FAQ — the sources engines re-read.
11A score you can recompute
Public arithmetic, not a black box
The integrity score is a formula you can check by hand: your accuracy ratio, minus five points per open critical, clamped to 0–100. It trends over 90 days by default, up to a year — and when there isn't enough history, it says so instead of drawing a line anyway.
- 100 × accurate statements ÷ total statements
- Minus 5 per critical · clamped 0–100
- "Not enough data for a trend" — shown verbatim, when true
score = 100 × accurate ÷ total
− 5 × each open critical
→ clamped 0–100Accuracy ratio — accurate statements over everything checked
Five points off for every critical still open
Floor 0, ceiling 100 — no negative drama, no extra credit
Integrity score trend
Your brand-integrity score over time (100 = no open issues).
Not enough data for a trend
Shown when history is short — never a fake line
12Unchecked is never "clean"
"Nothing found" and "never looked" are different facts
A dashboard that shows a calm green zero for a site it never scanned is lying politely. Brand Integrity Monitoring keeps three states apart — checked and clear, not enabled yet, and nothing to trend — and never lets one impersonate another.
- Checked and clear says so — in those words
- Not enabled says exactly that, never "zero issues"
- An unscanned brand is never displayed as accurate
No integrity issues — your brand is being represented accurately.
Checked · and clear
Integrity scanning is not enabled for this site yet.
Not checked — and says so
No scan history yet.
Nothing to trend yet
13No extra engine calls, fail-closed
It rides the answers you already collected
When enabled, the integrity check re-reads the answers your AI Visibility runs already stored — it never probes an engine again. It runs inside the existing run, savepoint-guarded so a judge failure can't break anything it rides on. And before it spends a cent, three gates all have to agree. If any gate says no, nothing is judged: it fails closed, never open.
- Re-reads the stored answer — no new engine call
- Savepoint-guarded inside the run it's attached to
- Judge cost metered at the published model price — a missing price throws, never records zero
No new engine call — the stored answer is re-read, not re-asked
Collected by your AI Visibility run. Re-read in place.
Compares each statement against your confirmed facts. Four verdicts.
Quoted verbatim · severity · suggested action.
Savepoint-guarded — a judge failure cannot break the run it rides on
Fail-closed — if any gate can't be satisfied, nothing is judged and nothing is spent
14What you can actually change
Fix the source. Watch the next check.
No tool can make a model retract a sentence — and this page won't pretend otherwise. What you can do is find the wrong claim, fix the source feeding it, and watch whether the answer moves on the next scheduled check. That cadence is what turns a correction from guesswork into a loop. The fact sheet is on every paid plan; automatic scanning and email alerts are on Pro and above — see pricing.
- Email on new critical issues — Pro and above, once scanning is enabled
- Each site keeps its own fact sheet and its own findings — one client's facts never become another's reference
- White-label client reports roll up criticals, warnings, and the engines affected
Finding
"$62 per seat" flagged wrong
Fix the source
fernwell.com/pricing updated
Next check
on the scheduled cadence
Claim moved
now quotes "from $24 per seat"
New critical integrity issue — Fernwell
ChatGPT: "Fernwell is SOC 2 certified." Your fact sheet forbids this claim. View the finding →
Email on new criticals · Pro and above · when scanning is enabled
White-label report · 1 critical · 2 warnings · engines: ChatGPT, Gemini
Brand Integrity Monitoring, answered.
You keep an editable fact sheet of the things AI should get right about your brand — what you do, who you serve, pricing, founding facts, positioning. SearchChamp drafts the first version from your own site and cites the page each fact came from, and your team confirms it. Checks then compare what AI engines say about your brand against that sheet and surface the exact claim an engine got wrong, what is actually true, and how serious it is.
Classic brand monitoring watches mentions — who said your name, where, and in what tone. This watches correctness: whether the claims AI engines make about your pricing, positioning, and capabilities actually match the truth. A mention tool tells you that you were talked about; this tells you that an engine told a buyer something about you that is wrong.
You do. The fact sheet is a single editable reference owned by your team, and only the facts you confirm become the standard a check is measured against. SearchChamp drafts it from your own site to save you the blank page, but nothing is treated as true until someone signs it off — and an out-of-date fact sheet produces confidently wrong findings, so it is worth treating like a source of truth your comms and product people both agree on.
Because the quote has to be verbatim. Every finding carries the exact sentence from the engine’s answer, and any finding whose quoted text is not an exact substring of that stored answer is discarded rather than repaired. A paraphrase never reaches you. On top of that, findings are graded four ways — accurate, outdated, wrong, or unverifiable — and “wrong” is reserved for a direct contradiction of something on your sheet.
Checks run on a scheduled cadence rather than continuously, because each one reads answers across every engine you track and those answers move day to day. That rhythm is what makes a correction verifiable — you can see a claim change between runs instead of wondering whether a single answer was a fluke.
Yes. Each site carries its own fact sheet and its own findings, so an agency or a multi-brand group keeps them fully separate — one client’s facts never become the reference for another. Findings also roll up into the white-label client report as critical and warning counts with the number of engines affected.
The brand fact sheet is available on every paid plan — write it, seed it from your site, confirm and edit it. Automatic scanning of AI answers against that fact sheet, and email alerts on new critical issues, are available on Pro and above. See the pricing page for what each plan costs.
No, and be sceptical of anyone who claims otherwise — no tool can edit a model’s output directly. What you can change is the evidence it draws on: engines reflect what your own site and the sources they trust currently say. So the workable path is to find the inaccuracy, fix it at the source that is feeding it, and watch whether the claim changes on the next check. Monitoring is what makes that loop possible rather than guesswork.
Keep AI honest about your brand
7-day free trial · credit card required · cancel anytime