Agents

Overview#

Agents run on a schedule and bring work to you. Everything else in Seerly answers a question you went looking for; an agent notices something and puts it in front of you without being asked.

There are four of them, each with a deliberately narrow scope. That narrowness is the point. An agent asked to watch everything produces the kind of output nobody reads after the third week, so each of these watches one thing and produces one kind of suggestion.

The four agents#

Trend Finder#

Scans social trends and surfaces content opportunities.

Trend Finder watches social platforms, specifically X/Twitter, Instagram and Reddit, looking for what is rising in your category right now. It produces trending keywords, content ideas, hashtags and opportunities, each one tied back to the post or thread it came from so that you can read the source yourself rather than take a summary on trust.

It deliberately ignores citation counts and visibility scores. It uses them in the background to work out where you are weak, and never reports on them, because that is the Visibility Analyst's job and keeping the two apart is what stops both agents producing the same vague weekly digest. It also remembers what it has already shown you, so the same thread does not resurface week after week.

Deploy this one first if you are deploying anything. The Brand Monitoring Scout produces noticeably better suggestions when it can read Trend Finder's output, and Seerly will nudge you about it if you deploy the Scout on its own.

Visibility Analyst#

Tracks your AI search visibility and competitor landscape.

Visibility Analyst watches AI platform visibility: citation health, competitor movements and platform scores. What comes out is detected changes in your citations, drops and gains with an attempt at the root cause behind each, competitor movements, and the recommendations that follow from them.

It only reports on the AI platforms your plan actually covers, which is worth knowing before you wonder why an engine is missing. It will not tell you that you are absent from an engine you do not track, since that would be noise rather than a finding.

Content Scout#

Proposes fresh content ideas from your team's research.

Content Scout turns findings into work, and it does no research of its own. It reads what the other agents have already found, along with your keyword universe, your topic and keyword gaps and your existing content, and proposes ideas from that material. The output is a shortlist of candidate content ideas, spread deliberately across different channels and keyword clusters, from which a few are selected. It lands in ActionsContentSuggestions, covered in Content suggestions.

Two constraints shape what you actually see. Every idea has to sit inside your keyword universe, which is what stops the Scout drifting into adjacent categories that are genuinely interesting and not yours. And ideas are checked against your existing content for near-duplicates by meaning rather than by title, so the same piece should not arrive twice wearing different words. Each idea comes through as a real editorial headline with a focus keyword behind it, rather than a keyword pasted into a title and called a brief.

Brand Monitoring Scout#

Recommends weekly prompt-tracking changes.

The Scout watches your tracked set itself, meaning the topics and prompts Seerly currently monitors, which makes it the agent responsible for keeping your measurement honest as the business moves underneath it. It produces prompts to add, drop or swap and topics to open or retire, landing in ActionsPromptsSuggestions. See Prompts.

It works across two kinds of prompt, and the distinction changes what you are being told. Discovery prompts ask whether you are found for an intent at all, which are your visibility questions. Sentiment prompts ask how you are talked about once you do come up, covering reviews, trust, quality, complaints and comparisons on experience.

The Scout treats a gap in sentiment coverage every bit as seriously as a gap in discovery coverage, which is worth knowing because most teams have never explicitly considered whether their reputation questions are being tracked. Its suggestions are usually where that surfaces for the first time.

It also proposes the set it believes is right rather than trimming to what it thinks you will accept, so you may well see more suggestions than you take. Declining several is the expected outcome rather than a sign anything is wrong.

The agent chain#

The four form a chain, and the order genuinely matters, because each one works from what the previous ones found.

The agent chain

1. 1

   Trend Finder

   Watches X, Instagram and Reddit for what is rising

   → findings the next agents read

2. 2

   Visibility Analyst

   Watches citations and competitor movement

   → findings the next agents read

3. 3

   Content Scout

   Reads what the first two found, plus your gaps and existing content

   → Content → Suggestions

4. 4

   Brand Monitoring Scout

   Watches the tracked prompts and topics themselves

   → Prompts → Suggestions

Deploying only the last agent in a chain gives you the weakest possible version of it, which is why Trend Finder first is the practical rule rather than a preference.

Deploying and running#

Each agent has a brief describing what it should watch and what it should care about, edited with Edit brief or Edit the brief.

The brief decides output quality more than anything else you control here, and it is worth more of your time than it usually gets. A vague brief produces suggestions that are technically relevant and practically useless. Naming the topics that matter commercially, along with the things you have consciously decided not to chase, is what turns an agent from a novelty into something whose output you act on.

Deploy Scout anyway appears when Seerly thinks conditions are not ideal but you want to proceed regardless, most often because Trend Finder has not been deployed.

Activity Log shows executions as they happen, with Agent executions will appear here as they run and No activity yet as the normal empty states on a newly deployed agent. Automatic runs are paused means scheduled execution has stopped, and Paused by Seerly means we paused it rather than you, which is worth reading rather than dismissing, since it usually points at something about the project's own data.

Activity Log

Active

an execution is running now

Completed

finished, output waiting for review

Failed

worth opening, the message says why

Manual

you triggered this one yourself

No activity yet

the normal state on a fresh deployment

Approving and rejecting#

Nothing an agent produces goes live by itself. Every output is reviewed, approved, edited or rejected by a person, and that is a design decision rather than a limitation waiting to be lifted.

Reject declines an output and Rejection reason captures why, with Proceed, Not now and Dismiss handling the rest.

Fill in the rejection reason every time. It is the mechanism by which an agent improves, it costs a few seconds, and "we do not sell into that market" shapes every future run in a way that a silent rejection cannot.

Telegram#

Connect Telegram and Connect now link a Telegram account so that agent output reaches you without opening the dashboard, and can be approved from your phone.

It is worth doing if you find output going stale before anybody looks at it. Agent findings are tied to a moment, and acting a week later on a position that has since shifted is often worse than not acting at all.

Best practices#

• Trend Finder deployed before the Scout, so the Scout has something to work from.
• A specific brief on each agent, naming what matters and what to ignore.
• Output you would have arrived at yourself, arriving sooner than you would have got there.
• A rejection rate that falls over time as your reasons accumulate.
• Quiet runs during quiet periods. An agent that always finds something is not necessarily finding anything.

Next steps#

1. Deploy Trend Finder first, then the others.
2. Write each brief properly before deploying rather than afterwards.
3. Review promptly, since agent output decays faster than most things in the product.
4. Always give a rejection reason.
5. Follow the output to where it lands: content ideas in Content Suggestions, prompt changes in Prompt Suggestions.
6. Connect Telegram if approvals are piling up unread.
7. Read any pause message in the Activity Log rather than redeploying blindly.

Limitations#

Agents work from your tracked topics, engines and prompts. They cannot see opportunities outside that set, so an agent's silence about an area you do not track is not reassurance about it.

Agents propose. They do not publish, change settings, or spend anything on your behalf.

The conversational agent is documented separately in Agent chat.