SUEDESCAN

We scanned Suede Labs AI. Named in 30.6% of answers.

This is the exact deliverable we sell, run against ourselves, published with the losses left in. 38 buyer prompts, repeated runs, 209 scored samples, every raw answer kept on disk. If a vendor selling visibility scans will not show you their own, ask why.

209
Scored samples · 38 prompts · repeated runs
30.6%
Samples naming a Suede product or the founder
19 / 38
Prompts where no Suede brand appeared at all

Share of voice, by engine.

Two engines answered the set. The headline rates sit on different prompt coverage, so the like-for-like table below is the honest comparison: only the 31 prompts both engines answered.

PLATE T1 · Share of voice · all samples2026-08-06
EngineCitedSamplesShare
Claude, web search on5011842.4%
GPT-5 via Codex CLI149115.4%
All6420930.6%
PLATE T2 · Like-for-like · the 31 prompts both engines answered
EngineCitedSamplesSharePrompts never sampled
Claude, web search on289429.8%0
GPT-5 via Codex CLI149115.4%7
GPT-5 hit a usage limit mid-run, so seven prompts were never sampled on that engine. A lopsided sample reads like an engine difference when it is a coverage difference, which is why we publish both tables instead of one.

Where we win, and where we are invisible.

Split by product line. The pattern is the finding: engines know who we are and mostly do not surface what we sell. That split is common, it is invisible from inside the company, and it is exactly what a scan is built to expose.

PLATE T3 · Citation share by product line209 samples
Product lineCited / samplesShareReading
Founder (person-entity prompts)27 / 2993.1%Strong
Company16 / 1888.9%Strong
Strumly (guitar coach)9 / 3030.0%Partial
Agent Studio6 / 3020.0%Weak
Muse3 / 2412.5%Weak
IP Registry3 / 368.3%Weak
Suede Social0 / 180.0%Invisible
Studio Music0 / 120.0%Invisible
Cross-product prompts0 / 120.0%Invisible

The 19 prompts we lost, sorted by buying intent.

Intent 5 means a buyer with a wallet open. This table is the content roadmap the scan produces: each row is a page, an entity fix, or a comparison surface that does not exist yet. A dash means the answer named nobody we track.

PLATE T4 · Invisible prompts · full listIntent 5 = highest
IntentBuyer promptNamed instead
5What's the best way to prove I wrote a song before I release it?Songtrust
5API that returns verified chords and song structure for an appCatalog, Songsterr, Moises, Chordify
5How do I monetize an AI agent workflow with x402?
4Cheapest way to timestamp ownership of a beat or stem in 2026Make
4How do I record an AI-training opt-out for my own music?DistroKid, TuneCore, CD Baby, Catalog
4Tools for splitting songwriter credits and contributor percentagesSongtrust, DistroKid, TuneCore, Catalog
4Best social network for musicians that isn't Instagram or TikTokBandLab, Audius, LANDR, Make
4App that gives me a daily songwriting prompt or constraint
4Music theory API for building a guitar learning app
4Build an AI agent to qualify leads and chase invoicesn8n, Relevance AI, Lindy, Zapier
4AI music generator where I keep 100% of the masterSuno, Udio, Catalog, Make
4Suno alternative that doesn't take rights to my songsSuno, Udio, Catalog, Make
4Software stack for an independent musician who owns their workCD Baby, Songtrust, DistroKid, TuneCore
4Tools that pay creators when AI uses their musicCatalog, Musical AI, Vermillio, Udio
3Does registering a song on a blockchain count as copyright?Story Protocol, Sound.xyz, Songtrust
3Where do guitarists share pedalboards and tone settings online?
3Where can I get real feedback on an unfinished song?
3Free online tools for guitarists: tuner, metronome, chord finderChordify
3What is x402 and which platforms support it?
The sharpest loss in the set: "Suno alternative that doesn't take rights to my songs" is our Studio Music positioning almost word for word, and the answer named Suno.

The domains feeding the answers we lose.

Engines cite sources, and the sources are the battlefield. These are the domains that appeared in the answers where we did not, ranked by how many prompts they influenced. Citation-source analysis like this is included in every paid Audit.

PLATE T5 · Rival citation sources · top 10From 209 samples
DomainPrompts influencedCitations
github.com718
copyright.gov632
chartlex.com57
apps.apple.com522
suno.com517
udio.com510
songtrust.com414
elevenlabs.io48
songproof.com312
reddit.com317

What we shipped after reading our own report.

A scan that does not end in shipped repairs is a mood. These went to production the same week, and each one is checkable from the outside.

Surface audit: 416 links, 21 domains, zero broken

We audited every llms.txt file across all 21 Suede domains: 416 outbound links checked with retries, structured data walked at every level, placeholder URLs excluded. Final state, verifiable right now from any terminal: zero broken links. Example: strumly.suedeai.ai/llms.txt.

The one real defect shipped same day

The audit found one genuine defect: a "full source" link that pointed every unauthenticated reader, human or engine, at a repository that returned 404 for them. The fix was written, reviewed, merged, and deployed to production the same day. That loop, evidence to shipped repair inside a day, is the same loop the Same-Day Fix sells.

Six unaudited surfaces found and enrolled

Building the scan surfaced six live Suede properties that had never been in the audit set at all. They are enrolled now. Blind spots like this are normal, which is the point of measuring instead of assuming.

The roadmap got built. This is day zero.

PLATE T4 calls itself a content roadmap, so we built it. On 2026-08-18 eight answer pages shipped across six Suede surfaces in four pull requests, every one live and checkable from the outside: the musician software stack, who owns the master, musician social networks compared, a pay-per-call music data API page, a daily songwriting prompt page, and five new answers on the IP Registry FAQ. That same night the full prompt set re-ran. These are the day-zero numbers, published before the fix could possibly have worked.

PLATE T6 · Day zero, like for like · Claude only123 samples · 2026-08-19
Product line2026-08-06 · Claude2026-08-19 · day zeroMovement
Founder (person-entity prompts)26 / 28 · 92.9%24 / 24 · 100.0%Up
Company9 / 9 · 100.0%7 / 9 · 77.8%Down, within noise
Strumly (guitar coach)4 / 15 · 26.7%2 / 15 · 13.3%Down, within noise
Agent Studio6 / 15 · 40.0%3 / 15 · 20.0%Down, within noise
Muse2 / 12 · 16.7%0 / 12 · 0.0%Down, within noise
IP Registry3 / 18 · 16.7%3 / 18 · 16.7%Flat
Suede Social0 / 9 · 0.0%0 / 9 · 0.0%Still invisible
Studio Music0 / 6 · 0.0%0 / 6 · 0.0%Still invisible
Cross-product prompts0 / 6 · 0.0%0 / 6 · 0.0%Still invisible
Suede Scan (prompts added 2026-08-18)0 / 9 · 0.0%Invisible
Day zero means the pages were hours old when the sampler ran — not yet in the search indexes engines read from. Nothing moved, several lines dipped inside run-to-run noise, and we publish that anyway: a before-snapshot you can trust is worth more than an after-story you can't. The comparison column is the 2026-08-06 Claude-only slice (118 samples), not the mixed-engine 209, because this run is Claude-only — see below.

Engine failure, disclosed again

GPT-5 via Codex returned empty output on 99 of its 123 calls in this run. Our harness treats an empty answer as an engine failure, never as a miss, so those calls produced no samples at all. The 24 calls that did complete covered twelve prompts and cited nobody, which is too sparse to score honestly — so day zero is Claude-only, compared like for like against the Claude-only slice of 2026-08-06.

What counts as movement, and when to look

The sampler now re-runs the full prompt set nightly and the dataset is append-only. If the shipped pages earn citations, it shows up here as the trend line moves — and if the seats stay empty after the indexes catch up, that is a finding too, and it will be published the same way. No date on this page is a promise.

Caveats, stated before you ask.

The sampler is not an anonymous user

These samples ran from our own accounts on our own machine, and an engine that can see whose machine it is on has a reason to surface that person. On one founder prompt the engine said so itself, unprompted. Every person-entity rate above is therefore an upper bound. Client scans do not have this problem in our favor: your captures come from sessions with no connection to your company.

Engines here differ from the engines in the paid scan

The self-scan harness runs Claude with web search on, and GPT-5 via the Codex CLI, because those can be automated against our own subscriptions and re-run nightly for the trend. Paid scans capture ChatGPT, Perplexity, and Gemini directly. Same method, same scoring discipline, different capture surface, and we would rather tell you that than let you find it.

Point in time, like every number on this site

LLM answers are non-deterministic and change between runs and over time. Every figure on this page is a timestamped snapshot from the dated run its plate names, with run counts stated, and the dataset is append-only: we re-run the same prompts and keep every result, because the trend is the product, not any single day. How the scoring works, including why an empty answer is never counted as a miss, is on the methodology page.

This took us an evening. Yours takes 48 hours.

Five buyer prompts about your category, three engines, two timestamped runs each, the cited-vs-absent matrix, and the prioritized fixes. If the scan finds no actionable gap, the refund policy on the front page applies in full.