What Little Collie Shepherd reads
What does Shepherd read? Only the comments on sources I connect?
Yes. Shepherd reads the public comment sections of the sources you connect — a YouTube channel, a WordPress site — and nothing else. On YouTube that means the comments on your public videos and Shorts. Comments on community posts, and the live chat during a premiere or stream, are not covered yet. Shepherd reads a comment once, when the comment arrives, to decide what to do with it.
Does Shepherd read my subscribers' profiles or history?
No. Shepherd keeps a record of comments and what you and Shepherd decided to do with them. Shepherd does not fetch profiles, subscriptions, watch history, or anything about a person beyond what they wrote to you.
Does Shepherd see comments on other people's channels?
No. Each account sees only its own connected sources. Nothing a commenter does on someone else's channel reaches yours, and nothing on yours reaches theirs.
What does connecting a source give Shepherd access to?
On YouTube, a scoped sign-in that lets Shepherd list comments on your channel and set their moderation status — remove, hold for review, and ban an author. On WordPress, a plugin that holds a comment at the moment it is posted, before it appears. Neither grants access to uploads, analytics, settings, or your account.
Can I disconnect a source, and what happens to what Shepherd already read?
Yes, at any time. With a revoked authority, reading comments and all moderation stops. What Shepherd already recorded stays in your account under the normal retention schedule. You can export all of your data or close your account at any time.
AI use
Does my comment get sent to an AI?
Sometimes. Shepherd reviews comments in stages. The first two — your banned words, patterns, and accounts, and a content normalization — run on Shepherd itself. Only comments that pass these first two are then analyzed by an AI. See the privacy policy for further details.
What information is sent to the AI?
The comment text is sent along with the title and a short excerpt of the video or post the comment was left on. We also optionally send the text of the comment being replied to for context. If you switch them on, a per-source flag that indicates that the content features people on camera and a short summary of recent activity on the source can be included. That's it!
What is never sent?
The commenter's name, channel or user id, email, IP address, avatar, and any history or score Shepherd holds about them. The model is not told whose comment it is reading.
Is anything used to train a model?
No. Along with our provider being barred from training on anything we send, we do not train models on your content or your audience's comments.
Does the rule-drafting tool send anything out?
Only when a moderator opens the tool, and only the example comments that moderator chooses to work with. The rule-drafting tool is human-initiated, and a separate path in our software from automated moderation, with its own controls.
What Shepherd does with a comment
What are the three stages, in plain terms?
First, your rules: words, patterns and accounts you have banned are matched before anything else looks at the comment. Second, the dressed-up repeats: the text is normalized so spaced-out letters and lookalike characters cannot walk a banned phrase past the first stage. Third, the hard ones: what is left is read in context by a model, and anything still unclear goes to your review queue.
What does "removed" mean on each platform?
On YouTube, the comment is set to rejected. This hides the comment from public view, but not from the author. You keep your engagement.
On WordPress, the comment is sent to the spam folder. The comment won't be displayed on your content at all.
Shepherd doesn't delete anything from your account and the platform; both are the platform's own moderation states, applied through the platform's own tools.
Can Shepherd ban an account, or only act on comments?
Both. A moderator can ban an author from the queue. On YouTube that uses the platform's own ban where the ban call works, and Shepherd's own rule where the call does not — either way the author's future comments are matched at the first stage. On WordPress, the rule is the ban.
Does Shepherd ever reply, like, pin, or post as me?
No, on purpose. Try as they might, AI cannot really replicate your voice. Not in the comments and not in your content. So, every reply from your account is written and sent by you or your moderation team.
Does Shepherd act on comments posted before I connected?
Nope! Reading starts the day you connect your account to YouTube or WordPress. On first connect, Shepherd offers a look back over the current day, which you run or not. At this time, we don't offer a look-back further than that.
How fast does Shepherd act on a new comment?
WordPress comments are acted on as they arrive. YouTube comments are polled regularly. When you post a new video on YouTube, Shepherd bumps up its protection with Debut Mode, running faster to keep up. If your video goes viral, Shepherd can turn on Alert Mode, running even faster, until comments quiet back down.
What happens when Shepherd is unsure?
Shepherd asks you. Any comment that Shepherd cannot categorize with confidence goes into a review queue with reasoning attached. Each of these queued comments stays live or held according to your settings until a person makes a decision. Shepherd never guesses.
Who sees the review queue, and can a moderator overrule Shepherd?
Everyone with a moderator or admin role on your account. A person's decision is final: keep, remove, or ban, and the queue records who decided and why.
Your rules
How are the rules set?
You set your policy for your content. Shepherd comes with a short list of slurs and patterns that are abuse in any channel. That starting point can be customized, even emptied entirely, to create your rules. What counts as abuse on your channel or site is yours to control.
Can I add words, phrases, and specific accounts?
Yes! You can block comments by words, patterns, phrases, and account names. Any account you ban is matched in the first stage, before any other rules are applied.
Can I trust people — regulars, friends, my own team?
Yes. Any account on your allow list goes straight through, untouched by any other review. You can also create a watch list for accounts you want to manually review every time, whatever they wrote.
Does Shepherd handle spaced letters, lookalike characters, and emoji substitutions?
Yes. Our second stage normalizes text, including Unicode lookalikes, extra spaces, repeated characters, and emoji. The results are then categorized against your policy's rules. So no h@xx0r or 🍆 escapes review.
Does Shepherd work in languages other than English?
Partially. Shepherd's starter list and pattern stage are written for English. Shepherd reads most languages and your own rules can be written in any language. If your audience comments are not mostly in English, let us know at [email protected]; we'd love to improve Shepherd's handling of them.
Can I set different rules per video or per post?
Not today. Rules apply across your account, to every connected source. Per-source settings are limited to a few switches, like the on-camera flag.
Mistakes and accountability
What if Shepherd removes something it shouldn't have? Can I restore it?
That depends on the platform. On WordPress, yes: a comment in the spam folder comes back with one click. On YouTube, a rejected comment cannot be un-rejected. Google does not allow for that. So, automated removal there stays off until you switch it on, and the review queue holds a comment rather than removing it when in doubt. You can also adjust how strict or lenient Shepherd is to fine tune its moderation for your audience.
Can I see why Shepherd made every decision?
Yes. Every decision is logged with the reason, in plain words, and with who or what made the decision — the first stage, the second, the model, or a named person. The log is yours to read and to export.
Will Shepherd remove criticism, blunt reviews, or disagreement with me?
No. Shepherd acts on conduct and how intense the conduct is, never on a viewpoint. Civil disagreement with you, however blunt, stays up, and there is no setting that changes that.
Does Shepherd remove comments for a viewpoint — politics, religion, anything like that?
No. There is no category for a viewpoint and no way to add one. The categories are kinds of conduct: slurs, threats, harassment, spam, and the like, at thresholds you set.
Does automated removal turn on by itself?
No. Until you approve a policy that turns automated removal on, Shepherd holds and queues but removes nothing unprompted. A person is in the loop on day one, and stays there until you decide otherwise.
Will a commenter know Shepherd removed their comment?
Shepherd never contacts a commenter and never labels a removal. What the commenter sees is the platform's own behavior for a removed or held comment, the same as if you had done the removal by hand.
Teams
Can more than one person moderate, and can I see who did what?
Yes. Invite moderators to your account; each signs in as themselves. Every action is recorded against the person who took it, so a shared queue does not become an anonymous one.
Can I give a moderator access without giving them my YouTube login?
Yes. You connect the channel once, as its owner. Moderators work in Shepherd, under their own accounts, and never see or hold your platform credentials.
Billing
What does "source" mean for billing?
One thing you connect: a YouTube channel, or a WordPress site. Shepherd sets its monthly and yearly price by the number of sources you connect. Two channels are two sources.
What if I add or remove a source mid-month?
Adding one is billed from that day, prorated on your next invoice. Removing one takes effect at the end of the period you have already paid for; there is no refund for the remainder.
How does the annual plan work?
Pay for a year and two months of it are free: twelve months for the price of ten, per source.
How does the referral month work?
Refer a creator who signs up and stays for three paid months, and your next month is free. Three consecutive months of service, paid; a proration from adding a source does not count as one.
Leaving, and your data
What do you keep, and for how long?
Your account — rules, decisions, log — while the account is active. Records about commenters do not wait for that: comment text, commenter names, and the reasons recorded against a decision are redacted 24 months after they were last relevant, and you can set a shorter period for your own account. Comments assessed as credible threats are kept for six years.
If I cancel, what happens to my rules, my log, my data?
You get an export window, then we delete account data on a fixed schedule. Billing records are kept as long as tax law requires, and nothing else is.
Can I export the decision log?
Yes. An account admin can export the whole account at any time, as a JSON file: comment text, commenter display names and platform ids, your allow, watch and deny lists with your notes, and the full moderation log.
Who at Little Collie can see my account?
Our staff can, but only do so to address support issues. Every access is recorded and we do not use your data for any other purpose. You and your comments are not the product; moderation is. See the privacy notice for further details.
Where is the data stored, and under what law?
In the United States, on our hosting provider, behind Cloudflare's network edge. Little Collie LLC is an Oregon company; the privacy notice sets out the transfer terms for customers in the EEA and UK.
Platforms
Which platforms today, and what is "coming soon"?
YouTube and WordPress today. Facebook and Instagram are being built and tested against the platforms' own rules; we will say a date when we have one.
Does Shepherd replace YouTube's built-in "held for review" and blocked-words settings?
Shepherd sits alongside them. YouTube's own settings keep working; Shepherd adds patterns YouTube's list cannot express, the normalized second pass, a longer read for the hard ones, and a log with reasons. Anything YouTube holds first, Shepherd never sees.
Will Shepherd put an AI label on my replies?
No. Shepherd never writes or sends a reply, so nothing Shepherd does is a machine reply that disclosure rules are written for. Those rules are about tools that answer your audience for you. Every reply from your account is yours. Recent changes in legislation are requiring such labels on generated content, in both images and text, depending on your jurisdiction.