Actions

Most AI visibility dashboards stop at the data. You read a chart, frown, close the tab, and go back to your day. Nothing actually changes.

Actions is the page that closes the loop. Everything Trakkr learns from your prompts, citations, crawler logs, perception data, and site audit feeds a single prioritized list of things you should do next, ranked by what will move your visibility the most. Some of those things are recommendations you read and decide on. Some are drafts Trakkr writes for you. Some are fixes Trakkr just ships, with your approval, into your CMS.

That's the trust spectrum, and it's the most important thing to understand about this page.

The trust spectrum, in one paragraph

Every action sits somewhere on a line. On one end: Trakkr suggests, you do the work. In the middle: Trakkr drafts, you review and send. On the other end: Trakkr executes, you approve and watch it happen. The same action page hosts all three modes because the right amount of automation depends on the action type, your connected integrations, and how much you trust us with what.

The spectrum exists because the work itself is uneven. Writing a comparison article that AI will actually cite is a one-shot, well-defined task that Trakkr can do end-to-end. Defending a rising competitor's narrative is a strategy decision, which is yours.

Where actions come from

Actions are generated, not curated. Three pipelines feed the queue.

The AI synthesizer. After each report run, Trakkr assembles a structured snapshot of your brand (visibility trend, top losing prompts, recent citation movements, competitor share-of-voice, crawler activity, perception shifts, site audit issues, traffic deltas) and hands it to Claude Sonnet 4.6 with a consulting-grade brief. The model returns up to four strategic actions plus a written briefing that ties them to the data. This is where the highest-impact, "you wouldn't have thought of it yourself" recommendations come from.

Rules-based emitters. Specific data signals create specific actions deterministically. A site audit issue becomes a fix_audit_issue. A new citation gap becomes a get_listed_on_source. A blocked AI crawler becomes an unblock_ai_crawlers. A losing prompt becomes a diagnose_prompt. These are the lights-on actions: they appear the moment the signal does, and they retire the moment the signal resolves.

Manual creation. Hit New action to add something Trakkr didn't catch. Useful for team-internal work you want tracked alongside the generated queue.

Two things are worth knowing about how this stays clean. First, every action carries a source_ref that makes it idempotent: if the same signal fires twice, the action updates rather than duplicates. Second, when the underlying signal resolves (the audit issue gets fixed elsewhere, the gap closes), Trakkr auto-retires the action. The queue is meant to reflect the world as it is right now, not a graveyard of stale tasks.

How to read an action

Every card carries the same shape. Top to bottom:

• A clear title (Fix meta descriptions on 12 pages, Pitch to publication: TechCrunch).
• A priority chip: quick-win, high-roi, major, easy, medium, low. This is computed from effort × impact, with bonuses for setup and technical fixes, and a demand multiplier when the action targets a high-volume prompt.
• A category: setup, content, technical, citation, competitive, optimization, reddit, diagnose. Used for filtering, nothing more.
• A first step: the concrete next thing to do, regardless of how automated the action is. "Open the diagnosis," "Review the drafted email," "Approve the CMS fix."
• An expandable detail with the data that triggered the action, the playbook, and any draft Trakkr has prepared.

When you open an action, you'll see one of three patterns depending on its automation tier:

The pattern is deliberate: even at the "executable" end, nothing goes out without your approval. The first click previews. The second click ships.

What Trakkr can actually do for you today

The honest answer: a lot, and growing weekly. Here's the current state of the spectrum, with concrete examples.

Trakkr executes (full automation). Trakkr drafts the content and pushes it to your CMS in one flow.

• create_content_for_gap, create_content_for_citations, create_faq_content, create_comparison_content, target_query_cluster, optimize_high_traffic_page, close_blind_spot. Each writes a real page, with schema, ready to publish.

Trakkr fixes your CMS (technical execution). Trakkr generates the fix and applies it through your CMS adapter (WordPress today, more coming).

• fix_meta_descriptions, add_schema_markup, add_llms_txt, fix_audit_issue, fix_model_gap, optimize_for_agents.

Trakkr drafts, you send (outreach and replies). Trakkr writes the artifact with full brand context. You review and send through your own channel.

• pitch_to_publication, get_listed_on_source, update_cited_page, improve_citation_coverage, draft_reddit_response.

Trakkr completes in-app (inline execution). One click and the action runs inside Trakkr.

• add_prompts, apply_tag_suggestions, diagnose_prompt, setup_reddit, schedule_audits, run_first_audit.

Trakkr guides you (deep link to the right page).

• connect_analytics, connect_cms, install_crawler_tracking, add_subreddit.

Trakkr suggests, you decide (manual today). These are the strategy actions, often the most consequential, where the work is yours.

• leverage_winning_position, protect_crown_jewel, defend_threatened_prompt, counter_rising_competitor, counter_narrative, launch_content_campaign, consolidate_pages, fix_model_crisis, unblock_ai_crawlers, investigate_traffic_drop, and a handful of crawler-budget rules that need robots.txt access we don't yet have.

The shape of this list changes every release. You can see the live state, with a per-type automation level, on the admin Actions Audit tab.

The lifecycle of an action

Every action moves through five states, and every transition is logged with the brand's visibility and presence at that moment. This is what powers the "did this action actually help?" view on completed items.

The states themselves:

Snooze is the right answer more often than dismiss. The data may shift, the action may matter again, and snooze keeps it warm. Dismiss is for "this recommendation is wrong for our business," not "not this week."

Measuring whether actions actually worked

Because every completion records the brand's visibility and presence at that moment, completed actions show up on the visibility chart as markers. Hover any marker to see the action title, the score at completion, and the score now. Patterns emerge quickly: the comparison page you shipped two weeks ago is now mentioned in 40% more recommendations. The crawler block you removed brought a 6-point lift in 10 days.

This is also why we resist the temptation to auto-complete actions. The whole feedback loop falls apart if the system marks itself done without you confirming the work shipped. The one exception is setup checklist actions, which auto-complete when Trakkr can verify the underlying state (analytics connected, audit run, crawler installed).

Handing actions to Automate

Actions tells you what to do. Automate (Workflows) decides what should happen without you in the loop.

The handoff lives on every action card. Open one, click Automate, and you get a flow that turns the recommendation into a watchable rule. "When this signal fires again, do this action automatically." The workflow runs on its own from then on; the corresponding action types stop hitting your queue because Automate is handling them.

A reasonable rule of thumb:

• First time a pattern shows up: let it land as an action, review carefully, ship by hand.
• Second time, same shape: start drafting the workflow.
• Third time: automate it. Your future self has better things to do.

The split between the two pages is intentional. Actions is for judgment. Automate is for everything you've already exercised judgment on.

Common questions

How are actions different from a to-do list?

A to-do list is whatever you remember to add. Actions is generated from real signal: visibility data, citation movements, audit findings, competitor moves, crawler logs. It also dies on its own when the underlying signal resolves, which a to-do list never does.

Will Trakkr really push changes to my live site?

Yes, for the action types in the "Trakkr fixes your CMS" tier, and only after you approve the preview. Every change is logged with a diff, a timestamp, and a one-click revert through your CMS change log. We don't push silently and we don't push without a connected, authenticated CMS.

What if the AI synthesizer suggests something that's wrong for my business?

Dismiss it. Once dismissed, the same recommendation won't re-fire unless the data picture changes meaningfully. If a whole class of suggestion keeps missing the mark, tell us through the agent or feedback link, the briefing prompt gets tuned against feedback patterns.

How often does the queue refresh?

The AI synthesizer runs after every completed report (typically daily). Rules-based emitters run on their own cadences: site audits, citation polling, crawler aggregation, competitor scans. Manual actions appear immediately.

Can I assign actions to teammates?

Yes, on team plans. Each action has an assignee selector. Assigned actions appear in the filtered "Mine" view for that person and trigger a notification.

Why is the action I just completed still showing on the dashboard chart?

Because completed actions are kept as historical markers, that's how the visibility-attribution view works. They disappear from your active queue but stay on the chart so you can see which actions correlated with score movement.

Can I create my own action types?

Not in the UI today. The action type registry is curated on purpose because each type has a specific generator, automation path, and draft template behind it. If you have a recurring task that needs its own type, send it through to the team, that's how new types make it into the registry.