> ## Documentation Index
> Fetch the complete documentation index at: https://docs.adclear.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Helping Adclear learn

> How your responses to comments train Adclear's agents, and how to raise issues with the right team.

## Overview

Every comment you respond to in Adclear does two things at once. It moves your promotion forward, and it teaches Adclear's agents what compliance looks like in your organisation.

Skip a response, leave a one-line reaction, or click through without context, and the agents do not learn. The mechanics of commenting are covered in the /marketing/overview and the /compliance/overview. This guide focuses on why how you respond matters, and how to make sure the right people hear about issues.

<Note>
  Throughout this guide we use uploader for the person uploading a promotion, and reviewer for the person assessing it. The two roles interact with comments differently, and that difference matters.
</Note>

## Before you start

You will need the right access to do any of this. The table below shows who can submit, who can review, and where to look for full role detail.

| You need to...                                                   | Minimum role                                                                            | Where it is covered  |
| ---------------------------------------------------------------- | --------------------------------------------------------------------------------------- | -------------------- |
| Upload a promotion and respond as the uploader                   | Marketing, Marketing Admin, Compliance, Compliance Admin, Legal, or External Influencer | /marketing/overview  |
| Review a promotion and respond as the reviewer                   | Compliance or Compliance Admin                                                          | /compliance/overview |
| Resolve an escalation                                            | Legal                                                                                   | /how-roles-work      |
| Tag @adclear-support on a comment                                | Any role with access to the promotion                                                   | This guide           |
| Open the Get Help bot to log a question, bug, or feature request | Any logged-in user                                                                      | This guide           |

<Tip>
  If you are unsure which role you have, check with your workspace admin. Full permission breakdowns are in /how-roles-work.
</Tip>

## Why your comments matter

Adclear is not a static rules engine. It is a set of agents that get sharper every time someone tells them whether a comment was right or wrong, and why. The "why" is the part most teams miss, and it is the part that actually trains the agents.

Three things happen every time you click a response button:

<CardGroup cols={3}>
  <Card title="The promotion moves forward" icon="arrow-right" color="#2563eb">
    Your response unblocks the next step. The reviewer can decide, the uploader can resubmit, or the asset can be approved.
  </Card>

  <Card title="The other person gets context" icon="message" color="#7c3aed">
    Uploaders see why a comment matters. Reviewers see why a dispute is reasonable. Without context, both sides have to guess.
  </Card>

  <Card title="The agents learn" icon="brain" color="#16a34a">
    Adclear's agents watch the response, the reasoning, and the outcome, then adjust how they comment on similar content in future.
  </Card>
</CardGroup>

### The agents are weighted toward reviewer responses

This is the most important thing to understand about how Adclear learns. When an uploader and a reviewer disagree on a comment, the reviewer's response carries more weight in how the agents adjust.

<Warning>
  For regulatory comments, only the reviewer's response trains the agents on what is compliant. An uploader disputing a regulatory comment without a reviewer agreeing will not change how Adclear comments on similar content in future. This is by design. The reviewer holds the foundation of what is compliant and what is not.
</Warning>

That has two practical consequences:

1. Uploaders cannot quietly silence a comment. Disputing a regulatory comment without supporting context simply sends it to the reviewer to decide. The reviewer's call is what counts.
2. Reviewers carry the responsibility for shaping the agents. A reviewer who clicks through with no rationale is not just slowing down the next reviewer. They are missing the chance to teach Adclear something it can reuse on every future promotion.

For tone-of-voice and non-regulatory comments the picture is more flexible. Patterns from uploader responses can feed into building guidelines over time, but the same principle holds: the more context, the better the learning.

## How each side responds

The three response options look similar between uploader and reviewer, but they mean different things, and the buttons are labelled differently for a reason.

<Tabs>
  <Tab title="Uploader responses">
    When you have uploaded a promotion and Adclear has commented on something, you have three options:

    | Button             | What it means                                                                     | What happens to the comment                                                           |
    | ------------------ | --------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------- |
    | Agree, Will Action | You agree with the comment and will make the change in the next version           | Comment moves to the Actioned tab. You are expected to address it before resubmitting |
    | Dispute            | You disagree with the comment and do not intend to change the promotion           | Comment stays visible with a Disputed status. The reviewer will see it and decide     |
    | Reply              | You are adding context or asking a question without changing the comment's status | Comment stays open. Useful for clarifying before deciding to action or dispute        |

    <Note>
      "Agree, Will Action" and "Dispute" both require substantive context. A one-line "yes" or "no" is not enough. See [What strong feedback looks like](#what-strong-feedback-looks-like) below.
    </Note>
  </Tab>

  <Tab title="Reviewer responses">
    When a submitted promotion comes to you for review, your three options are different:

    | Button                 | What it means                                                                              | What happens to the comment                                                           |
    | ---------------------- | ------------------------------------------------------------------------------------------ | ------------------------------------------------------------------------------------- |
    | Agree, Please Action   | You agree with Adclear and you are telling the uploader to make the change                 | Comment moves to the Actioned tab. The uploader must address it in the next version   |
    | Relevant, Not Required | The comment is relevant in general, but in this specific context it does not need a change | Promotion can be approved as-is. Adclear learns there is a nuance the agents missed   |
    | Dismiss                | The comment is wrong. Adclear has misread, misapplied a rule, or got the context wrong     | Comment moves to the Dismissed tab. This is an outright rejection of the agents' call |

    <Tip>
      The difference between Relevant, Not Required and Dismiss matters more than it looks. "Relevant, Not Required" tells the agents "the comment was sensible, but this context is an exception." "Dismiss" tells the agents "you were wrong here." They train the agents in different directions, so pick the right one.
    </Tip>
  </Tab>
</Tabs>

## What strong feedback looks like

Before getting into specific examples, it helps to know what a strong response is doing. Whether you are an uploader or a reviewer, the same eight dimensions apply. A response that hits more of these is one that the agents can actually learn from, and that the next person reading the comment can rely on.

| Dimension                | What a strong response does                                                           | Why it helps the agents learn                                      |
| ------------------------ | ------------------------------------------------------------------------------------- | ------------------------------------------------------------------ |
| Justified disagreement   | Explains why a comment is incorrect or over-applied                                   | Teaches boundary calibration and reduces false positives in future |
| Contextual awareness     | References audience, timing, channel, or market norms                                 | Helps agents learn situational interpretation                      |
| Regulatory grounding     | Anchors feedback to identifiable standards or principles (such as COBS, FCA Handbook) | Improves rule-mapping precision                                    |
| Harm reasoning           | Explains how (or why not) consumers could be misled or harmed                         | Strengthens risk modelling                                         |
| Proportionality          | Distinguishes material issues from immaterial ones                                    | Prevents over-enforcement on minor concerns                        |
| Completeness             | Identifies missing required information                                               | Helps agents understand where gaps in a rule exist                 |
| Constructive remediation | Suggests what should change or be added                                               | Enables corrective learning rather than just rejection             |
| Nuanced judgement        | Recognises grey areas or uncertainty                                                  | Encourages probabilistic reasoning instead of absolute calls       |

<Tip>
  You do not need to hit every dimension in every comment. The point is to recognise that a substantive response usually covers at least three or four. A response that hits none of them, regardless of which button you clicked, is a missed signal.
</Tip>

### Strong feedback in practice

Compare the two columns below. Same comment, two very different signals to the agents.

<CardGroup cols={2}>
  <Card title="Weak (do not do this)" icon="circle-xmark" color="#dc2626">
    Uploader disputes: "disagree, not needed"Reviewer dismisses: "not required"

    Neither of these tell anyone, human or agent, why. The agents cannot generalise from them, and the next person to read the comment has to start from scratch.
  </Card>

  <Card title="Strong (do this)" icon="circle-check" color="#16a34a">
    Uploader disputes: "Our latest tone of voice guidelines allow 'easy to utilise' in the context of our mobile app, not in terms of ease of investing. Asset stays as is."Reviewer dismisses: "Dismissed. The phrase refers to in-app features being utilised, not investment products. In line with our tone-of-voice agreement on being clear and not misleading, the statement remains."Reviewer agrees: "Agree, please action" is enough on its own. The Adclear comment already explains the rule, you do not need to repeat it. Brevity is fine here.
  </Card>
</CardGroup>

## What goes wrong without context

These are the failure modes we see most often. Each one has a hidden cost for the team, for the audit trail, and for what Adclear's agents are able to learn.

<AccordionGroup>
  <Accordion title="1. Uploader disputes with no rationale, so the reviewer cannot tell whether to push back or let it go" icon="user">
    The Adclear comment: "The word 'easy' may trivialise the nature of investing, which carries inherent risk. Consider revising in line with COBS 4.2.1R requirement that communications are fair, clear and not misleading."What the uploader does: Clicks Dispute and writes "not needed."The cost:

    * The reviewer opens the promotion and sees a disputed comment with no reasoning. They now have to either guess what the uploader meant, risking approving something they should not, or kick it back with a question, adding a full review cycle.
    * The agents see a dispute with no signal and learn nothing. The next promotion with the same comment will trigger the same back and forth.
    * The audit trail shows a disputed compliance comment with no documented justification.

    Better: "Disputed. The word 'easy' here refers to the in-app onboarding flow, not the act of investing. The caption clarifies it is the navigation that is easy, not investment outcomes. Per our tone-of-voice policy section 3.2, we use 'easy' for product usability but never for investment ease or returns. Asset stays as is."
  </Accordion>

  <Accordion title="2. Reviewer dismisses with no rationale, so the agents are told they are wrong but not how to be right" icon="circle-exclamation">
    The Adclear comment: "The word 'utilise' may not comply with tone-of-voice guidelines requiring simple, accessible language."What the reviewer does: Clicks Dismiss and writes "not applicable."The cost:

    * The uploader sees a dismissed comment and assumes Adclear is unreliable on this kind of issue. They start to ignore similar future comments themselves rather than addressing them, and the tool's credibility erodes for them.
    * The agents learn "this kind of comment was wrong" but have no signal about why, so they may swing the other way and stop commenting on similar content even where they should. The reviewer holds the foundational knowledge of what is compliant. If they do not share their reasoning, the agents cannot absorb it.
    * No defensible record exists of why the dismissal was correct.

    Better: "Dismissed. The agents have read 'utilise' as a tone-of-voice issue, but the word does not appear in the asset itself. It only appears in the caption, which is internal context for review and is not customer-facing. The tone-of-voice rule does not apply."
  </Accordion>
</AccordionGroup>

## When the comment itself is the problem: Notifying Adclear

Sometimes the issue is not whether a comment is right or wrong. It is that Adclear has misread the promotion entirely, missed something it should have caught, or hit a bug in the tool. You have three ways to notify the Adclear team, and each is for a different situation.

### Tag @adclear-support on a specific comment

Use this when Adclear has commented on something incorrectly. For example, the agents are misreading a word, applying a rule that does not fit, or referencing content that is not actually in the asset.

<Steps>
  <Step title="Open the comment thread">
    Reply to the relevant Adclear AI comment from the right-hand panel.
  </Step>

  <Step title="Tag support and explain">
    Use the @ symbol to tag `@adclear-support` and add specific context. For example: "This is misreading the promotion, the word 'utilise' does not actually appear in the asset."
  </Step>

  <Step title="Submit the comment">
    The Adclear team is notified against this specific promotion and a ticket is created.
  </Step>
</Steps>

### Add a sticky comment for something Adclear missed

Use this when Adclear has missed a feature or element of the promotion that should have been raised. No existing comment exists for you to reply to.

<Steps>
  <Step title="Click the + comment icon (top right of the asset viewer)">
    This is the sticky-comment tool. It lets you place a marker anywhere on the asset.
  </Step>

  <Step title="Drop the marker on the relevant element">
    For example, on a rocket image in an investments ad: "Rockets should not be used here as they trivialise investment risk."
  </Step>

  <Step title="Tag support">
    Tagging `@adclear-support` against the new comment notifies the Adclear team. We then review the agents and train them so this gets picked up automatically in future.
  </Step>
</Steps>

### Use the Get Help bot for questions, bugs, and feature requests

Use this for anything not specific to a single comment. General questions, suspected bugs in the tool, or feature ideas.

The Get Help button is in the top right of any promotion (and also via the chat icon in the bottom left). It opens the Adclear support bot, which categorises your message into one of three flows:

<CardGroup cols={3}>
  <Card title="I have a question" icon="question">
    For how-do-I and why-is-this questions. Quickest path to an answer.
  </Card>

  <Card title="I found a bug" icon="bug">
    For broken behaviour, errors, or things not working as documented.
  </Card>

  <Card title="I want a new feature" icon="lightbulb">
    For ideas and improvements. These feed into the product roadmap.
  </Card>
</CardGroup>

<Tip>
  At any point you can ask to speak to a human. Just type "speak to an agent." We monitor every comment that comes through the bot, and the bot exists to help us spot themes and respond consistently, not to keep you at arm's length.
</Tip>

## Quick reference

A printable summary of who does what.

<Tabs>
  <Tab title="If you are an uploader">
    | When you...                                | Click                     | And add...                                  |
    | ------------------------------------------ | ------------------------- | ------------------------------------------- |
    | Agree with the comment and will fix it     | Agree, Will Action        | What you are changing and why               |
    | Disagree and will not change the promotion | Dispute                   | The specific reason: context, rule, policy  |
    | Need clarification before deciding         | Reply                     | The question you need answered              |
    | Spot Adclear has misread something         | (any reply)               | Tag @adclear-support with what it got wrong |
    | Spot something Adclear missed              | Sticky comment via + icon | Where, what, and tag @adclear-support       |
    | Have a general question or bug             | Get Help button           | Pick the right category in the bot          |

    Resubmit only when every comment has a status applied. Leaving a comment unaddressed means the reviewer has to chase it.
  </Tab>

  <Tab title="If you are a reviewer">
    | When you...                                              | Click                     | And add...                                                                       |
    | -------------------------------------------------------- | ------------------------- | -------------------------------------------------------------------------------- |
    | Agree the comment is right and the uploader must fix it  | Agree, Please Action      | A short acknowledgement is enough. The Adclear comment already explains the rule |
    | Comment is reasonable but does not apply in this context | Relevant, Not Required    | The specific nuance that makes this an exception                                 |
    | Comment is wrong                                         | Dismiss                   | A prescriptive reason: what the agents got wrong, and what rule actually applies |
    | Spot Adclear has misread something                       | (any reply)               | Tag @adclear-support with what it got wrong                                      |
    | Spot something Adclear missed                            | Sticky comment via + icon | Where, what, and tag @adclear-support                                            |
    | Need legal or specialist input                           | Final review action       | Use Escalate and select the right person (where enabled)                         |
    | Have a general question or bug                           | Get Help button           | Pick the right category in the bot                                               |

    Respond to every comment before clicking Review. A skipped comment is a missed training signal, and the agents are weighted toward your responses, so what you skip, they do not learn.
  </Tab>
</Tabs>

## Frequently asked questions

<AccordionGroup>
  <Accordion title="If I am an uploader and I dispute a comment, do the agents learn from that?" icon="circle-question">
    For tone-of-voice and non-regulatory comments, yes. Uploader responses contribute, especially when patterns emerge. For regulatory comments, the agents are weighted toward the reviewer's response. An uploader dispute alone will not change how Adclear comments on similar content. The reviewer's agreement or dismissal carries the regulatory training signal.
  </Accordion>

  <Accordion title="Why does 'Relevant, Not Required' exist if I can just dismiss?" icon="circle-question">
    They train the agents in opposite directions. Dismiss says "the comment was wrong." Relevant, Not Required says "the comment was sensible, but this context is an exception, please learn the exception." If you dismiss comments that were actually relevant, the agents will eventually stop commenting on similar content even when they should, because you have told them the comment itself was incorrect.
  </Accordion>

  <Accordion title="What is the minimum I need to write in a comment?" icon="circle-question">
    For uploader disputes and reviewer dismissals: enough that someone reading it next week, without you in the room, can understand both the decision and why. Name the context (what the promotion is, what channel, what product), and name the rule or policy that justifies the call.

    For reviewer agreements: a short acknowledgement is fine, since the Adclear comment already explains the rule. You do not need to repeat the regulatory citation when you are agreeing with it.

    See the section above for the full set of dimensions.
  </Accordion>

  <Accordion title="What is the difference between tagging @adclear-support and using the Get Help bot?" icon="circle-question">
    `@adclear-support` ties your message to a specific promotion and a specific comment. Use it when the issue is about how Adclear commented (or did not comment) on something on that asset. The Get Help bot is for everything else: general questions, suspected bugs across the tool, or feature requests. Both notify the Adclear team. The difference is whether the context is the asset or the tool.
  </Accordion>

  <Accordion title="Can I email Adclear directly instead of using the bot?" icon="circle-question">
    The bot and `@adclear-support` are more efficient for us to triage. When you raise something through either channel, the context, user information, and the specific promotion or comment are pinned directly to the request, so we can pick it up and act without needing to chase down the background. Direct emails strip that context and slow the response down.
  </Accordion>

  <Accordion title="What if I disagree with the reviewer's decision after they have responded?" icon="circle-question">
    The reviewer's decision ultimately stands. They hold the keys to your organisation's policies and are the foundational source of what is compliant in your context. Disagreements should usually be resolved through comments and conversation on the promotion itself, not by overriding the decision.
  </Accordion>

  <Accordion title="When does escalation come into play, then?" icon="circle-question">
    Escalation is by exception, not a routine route. It is used when there is genuine grey area that needs legal or specialist input, or where an Uploader user with Risk Accepting permissions needs to formally accept risk on a specific change request. It is not a way to challenge a reviewer's standard call.

    The Escalate button is also not enabled for every organisation. If your workspace does not show it, escalations are not part of your configured workflow. Speak to your workspace admin or contact Adclear support if you think it should be enabled.
  </Accordion>
</AccordionGroup>

## Related guides

<CardGroup cols={2}>
  <Card title="Marketing Guide" icon="bullhorn" href="/marketing/overview">
    Full mechanics of uploading, choosing the right upload type, and submitting promotions for review.
  </Card>

  <Card title="Compliance User Guide" icon="clipboard-check" href="/compliance/overview">
    Full mechanics of reviewing promotions, making decisions, and handling escalations.
  </Card>

  <Card title="How Roles Work" icon="users" href="/how-roles-work">
    Permissions, role definitions, and how to choose the right role for a team member.
  </Card>

  <Card title="Promotion Statuses" icon="circle-info" href="/promotion-statuses">
    Every status a promotion can be in and who acts next at each stage.
  </Card>
</CardGroup>

## Still need help?

Contact support at [support@adclear.ai](mailto:support@adclear.ai).