Ask your Search Console
what happened
Search Console will tell you clicks are down 21%. It will not tell you why. Connect it to Hi, Moose and ask in plain language. Moose pulls the rows, cuts them the way an analyst would, and answers with a cause and the evidence under it.
SamPremium
Sam, looking across your Google Search Console performance data for the last two months, the drop in Google traffic is driven primarily by planting season seasonality, specifically the close of the spring planting window in late June.
Brand queries remain steady, but non-branded search volume for planting terms dropped sharply as the season moved past transplanting.
Comparing the prior 28 days against the recent 28 days:
Last year spring planting terms alone accounted for a large portion of your best-performing keywords.
- Spring planting queries fell 61% between the two periods, with the window closing in late June.
- Brand queries held within 2%, so this is demand moving, not a ranking loss.
- Fall planting and cover crop queries are climbing, two weeks behind last year's curve.
Chief desk supervisor. Believes every traffic drop has a simple explanation, usually a squirrel. Occasionally right.
A line went down. Now explain it to your boss.
Search Console is a great record and a poor explanation. It shows you the shape of the fall. Working out the cause means comparing periods, splitting branded from non-branded, checking whether position moved or demand did, and knowing enough about the business to recognise a season when you see one.
That is an hour of pivot tables for a question you will be asked again next month. Moose reads the same rows, runs the same cuts, and writes the answer with the numbers attached, so you can check the reasoning instead of rebuilding it.
Clicks, last 28 days versus previous 28 days. That is the whole message. The rest is yours to work out.
The spring planting window closed in late June. Brand queries held. Non-branded planting queries fell 47%. Average position moved 0.5, which is noise, not a ranking loss.
Questions that used to need a pivot table
Every one of these is a query plan Moose builds for you: date ranges, dimensions, filters, comparisons. You type the question.
The way most people do this
Export, compare, filter, repeat. None of it is hard. All of it is slow, and none of it is saved anywhere the next person can find.
The UI export truncates, and the truncation lands exactly on the long tail where a seasonal shift shows up first. You do not see what you are missing.
A meaningful slice of clicks sit behind queries Google will not name. Totals will not reconcile with the query table, and any analysis that ignores that is quietly wrong.
Branded/non-branded is a rule someone wrote once. It misses misspellings, new product names, and competitor brand terms, and nobody notices for months.
The reasoning lives in a spreadsheet on someone's laptop. Next quarter, the same question gets answered from scratch, differently.
Connect once, then just ask
Search Console connects through your own Google account. The rows come down to your machine, and the reasoning happens against real data, not a summary of it.
One OAuth screen under Connections. Read-only scope. Pick the property you want and Moose ties it to that project.
Moose turns the question into the queries an analyst would run, pulls the rows it needs, and keeps pulling until it can answer the question you asked.
The chat stays in your library with the tables intact. Share it, export it, or ask the same question again next month and compare.
Moose talks to the Search Console API from your computer with your own credentials. The rows land in local storage, not in a Hi, Moose database. If you run a local model, the analysis never leaves the machine at all.
The cuts a good analyst makes before answering
"Traffic is down" has maybe six real causes. Moose rules them in or out one at a time and tells you which cut carried the answer.
Impressions falling with position flat is demand. Impressions flat with clicks falling is a snippet or a competitor above you. The two look identical on the headline chart and need opposite responses.
Brand demand tracks your marketing. Non-brand tracks your SEO. Moose splits them and reports each separately, because a drop in one and not the other is most of the answer.
A 21% drop is rarely spread evenly. Usually a handful of query clusters or pages carry nearly all of it. Moose ranks by absolute click loss, so you look at the rows that matter.
A cliff on one date points at an update, a migration or a deploy. A slope over three weeks points at seasonality or slow decay. The shape of the decline narrows the causes fast.
The most common cause of a scary chart is the calendar. Moose pulls the same weeks from last year and tells you whether this is a repeat, with the year-over-year numbers next to it.
Queries that had impressions last period and none this one. Sometimes noise. Sometimes a page that fell out of the index and nobody noticed.
The answer is a conversation, not a report
A static dashboard makes you formulate the next question somewhere else. Here you just keep going, and the context comes with you.
Questions people ask
Is it making the numbers up?
No. Every figure in an answer comes from a Search Console API response Moose received during that chat, and the tables are printed from those rows. If a query returns nothing, Moose says so rather than estimating. That rule is the whole point: the product does not fabricate evidence, and a plausible number is worse than no number.
How does it know my business well enough to say "seasonality"?
From the query data and from your Context. Moose sees which query clusters fell and when, and Context holds what you have told it about the business: what you sell, who buys, what your calendar looks like. A drop that lands the week a season ends, on plant-name queries only, with brand queries flat, has one obvious reading. Moose names it and shows the cut, so you can disagree.
What about the 16 month limit and data delay?
Both are Google's limits and both apply. Search Console holds 16 months, and the last two or three days are usually incomplete, so a year-over-year question works and a three-year one does not. Moose respects the boundaries and says which dates it used. Every table names its date range for that reason.
Does this work with a local model?
Yes, and it is the reason some people run Moose at all. A local model means your query data never leaves the machine. The tradeoff is honest: a small local model handles the pulls and comparisons well, and a frontier model writes a sharper explanation of what it found. Switch per chat from the model picker.
Can I run this for several sites?
Yes. Each project holds its own property, its own context and its own chat history, and you switch with the project picker at the top of the sidebar. Agencies tend to run one project per client and ask the same three questions on the first of the month.
Do I need a paid plan?
No. Search Console chat runs on BYOK Free with your own key or a local model. The connection is your Google account and the compute is yours, so there is no meter on it.
More things to ask Moose
All use casesThe searches Google writes for itself before it answers, and the domains it cites.
Twelve searches behind one prompt, including the site: queries naming brands it trusts.
One question, five real engines, about a minute. Free on every plan.
A semantic index of every page, built and kept on your computer.
Share drops past your threshold, Moose finds the page, prepares the fix, waits for you.
The full list, and what Moose does step by step in each one.
Stop exporting.
Start asking.
Download Hi, Moose, connect Search Console, and ask it the question you were going to spend an hour on.