The Weekly SEO Workflow yeet Runs So You Don't Have To

The manual version of this job looks like: check Search Console on Monday, check rankings mid-week if you remember, glance at backlinks whenever a tool sends a scary email, and never quite get around to People Also Ask or checking whether ChatGPT knows you exist. yeet runs all of it on a schedule, whether or not you log in.

The watchers, and what each one is looking for

Watcher	Cadence	What it's looking for
GSC analytics	every pipeline tick (~15 min)	High-impression, low-CTR pages — ranking fine, not earning the click
Rank tracking	daily	Sustained position drops, filtered from routine noise
Backlink snapshots	daily	Links you had and lost since the last snapshot
Competitor tracking	daily	Movement from the sites you're actually being compared against
People-Also-Ask mining	rolled into the investigation cycle	Question gaps — things people ask that you don't have a page answering
AI-citation checks	~8 hours	Whether Perplexity-style answers are citing you
Prompt-rank tracking	~12 hours	Where you land across tracked AI prompts
LLM-mention metrics	~12 hours	ChatGPT and Google AI Overview mention probes (Perplexity)

From raw signal to a card

None of the above writes anything to your site directly. Watchers produce findings; findings get bundled into an investigation, and yeet runs investigations every two hours whenever there's new material to look at — no point running one against an empty inbox. An investigation is where yeet reasons about a finding: is this a real gap, is there a plausible fix, what fix type does it map to, and what's a defensible prediction for what fixing it should move.

Out of that comes an optimization card — a fix scoped to one URL or one issue, with a prediction attached. What happens to that card next depends on your autonomy mode: it sits in your queue, auto-opens as a PR, or, rarely, auto-merges. Underneath all of that, a verifier sweep runs every 30 minutes checking whether cards already in flight have moved the metrics they predicted — covered in full in how yeet verifies fixes.

What each watcher catches, in more detail

GSC analytics: the low-CTR pile

This is usually the highest-volume source of cards, because most sites have more "ranking fine, nobody's clicking" pages than they realize. yeet flags pages where impressions are meaningfully higher than the click-through rate would predict — the signature of a title tag or meta description that isn't doing its job at the position you're actually ranking in.

Rank tracking: drops, not noise

Rankings wobble constantly for reasons that have nothing to do with your page — a SERP feature rotated in, Google tested a variant, a competitor briefly outranked you and slid back. Daily rank tracking with drop detection is built to separate a real, sustained slide from that kind of noise, so you're not chasing every daily fluctuation.

Backlink snapshots: what you lost, not just what you have

A backlink count on its own doesn't tell you much. The daily snapshot exists to catch links you had and lost — a partner site redesigned and dropped the link, a directory delisted you, a guest post got pulled — because a lost link is a far more actionable signal than a static count.

Competitor tracking

Same daily cadence, watching the sites you're actually being compared against in SERPs you care about, so a competitor's content push or new page shows up as a signal rather than something you notice three months later when you happen to search the keyword yourself.

People-Also-Ask mining

PAA boxes are a fairly direct read on what real people ask around your topic. Mining them for questions you don't currently answer is one of the more reliable ways to find content gaps worth filling, as opposed to keyword-tool suggestions with no evidence anyone actually asks them.

AI-citation checks, prompt ranks, and LLM-mention data

These three are the AI-visibility side: Perplexity-based citation checks roughly every eight hours, prompt-rank tracking roughly every twelve hours, and LLM-mention probes covering ChatGPT-style and Google AI Overview prompts on a similar twelve-hour cadence, plus GA4 tracking of referral traffic from AI sources. Together they answer a different question than classic rank tracking: not "where do you rank" but "are you showing up in the answer at all."

What's free, what's paid

GSC analytics runs off your own Search Console connection, so there's no meaningful data cost attached to it. Rank checks, competitor discovery, and AI-visibility watchers lean on Exa, Perplexity, and Firecrawl rather than a classic SEO data vendor — yeet still budgets API spend carefully so research stays sustainable across plans. In practice this mostly shows up as a ceiling on how many sites and keywords get deep, frequent tracking rather than a hard wall — see the pricing page for how site counts and daily submission caps scale across Free, Pro, and Agent.

What this replaces, and what it doesn't

This system replaces the habit of manually checking four or five dashboards on some inconsistent schedule. It doesn't replace judgment about which findings actually matter to your business — that's still what the card-review step is for, whether that's propose_only review or PR review under auto-apply. See the credits guide for what each of these actions costs, and the docs for connecting Search Console and GitHub so the whole loop can run.

Quick FAQ

Do I need to do anything for these watchers to run? Just connect Search Console, and GitHub if you want auto-apply. Once connected, the pipeline tick picks up new GSC data automatically.

Why does rank tracking run daily instead of continuously? Rankings don't move meaningfully hour to hour, and continuous polling on paid rank data would burn through the per-user data budget for no real benefit.

What happens if a watcher finds nothing? Nothing — no investigation runs, no card gets created. Investigations only fire when there's new material worth reasoning about.