Reading your analytics with AI is a one-sitting job: export three reports from Google Analytics and Search Console, paste one brief, and get back a short list of what to do next. The shape is Skill and Tech, and the whole read takes about an hour, most of it spent checking what the AI says against the files. This page covers reading Google Analytics and Search Console exports with AI. If your question is whether your social media pays for itself, our guide to the ROI of social media for a local business covers that. The tools collect the data. The AI reads the export. You decide what is worth doing.
- Shape
- SkillTech
- Sitting
- one sitting
The shape: Skill and Tech
This job is a Skill because the value is in the reading, not the tool. Anyone can open a report. The hard part is knowing which numbers matter, comparing two periods fairly and turning a change into an action. A good brief carries that skill for you, and the AI applies it to rows you would never sort by hand.
It is also Tech because you have to get the data out first. Google Analytics and Search Console both have export buttons, and both hide them in slightly different places. You need a Google account with access to both properties, and Search Console has to be set up for your site already. If Search Console is not set up, that is the first job, and the read waits until it has collected a few weeks of data.
What that means for you: budget ten minutes to export, fifteen to paste and wait, and the rest of the hour to check. After the first time, the exports take five minutes because you know where the buttons are.
What you give the AI
Give it three exports, each covering the last 28 days compared with the 28 days before. If your business is seasonal, also export the same 28 days from last year, because a slow August against a busy July tells you nothing.
- Search Console, Performance report, Queries tab. In Search Console, open Performance, then Search results, set the date range to compare, and use the Export button. This gives you your top searches, up to 1,000 rows, with clicks, impressions, click-through rate and average position.
- Search Console, Performance report, Pages tab. You already have it: the CSV export downloads as one zip file with a Queries file and a Pages file inside. Unzip it and attach those two files.
- Google Analytics, landing page report. In Google Analytics, open Reports, then Engagement, then Landing page. Set the same comparison dates and download the file as a CSV from the share menu. If you track bookings or form sends as key events, they appear as a column in this report.
Add three short facts the files cannot hold:
- What your business sells and where, in one sentence.
- What a good visit looks like for you: a phone call, a booking, a contact form or a purchase.
- Anything that changed on your site or in your business during either period, such as a new page, a sale, a closure or a redesign.
Before you upload, open each file and check two things. The dates match the range you meant to export. No column holds personal information such as names or email addresses. Standard exports from these reports do not, but a custom report someone built for you might.
What comes back
Paste this brief into a new chat, then attach the files.
You are reading website analytics for a small business.
Business: [one sentence: what you sell and where].
A good visit is: [a call, a booking, a form, a purchase].
What changed during these periods: [new pages, sales, closures, a redesign, or "nothing"].
Attached:
1. Search Console queries export, last 28 days vs the 28 days before.
2. Search Console pages export, same dates.
3. Google Analytics landing page export, same dates.
Do this in order:
1. Confirm the date ranges in each file and say if any file looks incomplete.
2. List the five queries with the biggest change in clicks, up or down, with the numbers.
3. List queries with high impressions and a low click-through rate, where the average position is between 4 and 15.
These are pages Google already shows but people skip.
4. List pages that get search clicks but lead to few of the visits I care about.
5. Give me no more than three actions for this month, ranked by likely effect.
For every claim, name the exact rows and numbers it comes from.
If the data cannot answer something, say so instead of guessing.
Do not recommend buying ads or tools.What comes back is a short report in five parts that follow the steps. The useful part is usually step three, the near misses. A query where you show up often, sit near the top of the results and still get few clicks usually points to a page title or description that does not match what the searcher wanted. That is a change you can make today.
The three actions should read like tasks. "Rewrite the title on the pricing page so it names the service and the city" is a task. "Improve your SEO" is not. If you get the second kind, reply with: "Make each action something I can finish in under an hour, and name the page."
How to judge it
A good read points at rows. Every number in the report should be one you can find in the file in under a minute. Spot-check three of them before you act on anything.
A bad result looks like one of these:
- Numbers you cannot find. The AI states a figure that is not in any file, or rounds a change into something bigger than it is. Ask for the row. If it cannot give one, discard the claim.
- Generic advice. "Post more consistently," "build backlinks" or "improve site speed" with no page named. That advice could apply to any site, so it came from training data, not from your files.
- A trend from too little data. A query that went from two clicks to six is not a trend. Anything under about twenty clicks in both periods is noise.
- Blaming the wrong thing. A drop in clicks during the week you were closed for a holiday is the holiday. The AI only knows about it if you told it.
- Mixing the two tools. Search Console clicks and Google Analytics sessions are different counts and will never match. A report that treats the gap as a problem has misread the data.
- More than three actions. A list of twelve recommendations is a sign the AI is listing everything instead of choosing. Ask it to pick three and say why.
If the report passes the spot-check and the actions name real pages, it is good enough to act on.
What stays with you
The AI reads numbers. It does not know why they moved.
You know that the spring rush started late this year, that a competitor closed down the street, that the booking page was broken for a week in the middle of the period. Those facts decide whether a drop is a problem or a calendar. Only you can add them.
You also decide what is worth doing. The AI may rank a page title rewrite first, but if that page sells your least profitable service, it belongs at the bottom of your list. Effort and margin are yours to weigh.
Some things the exports cannot show at all. Phone calls from people who found you on Google and never visited the site do not appear in Google Analytics unless you track them. Visitors who decline cookies or block tracking may be missing from Google Analytics. Word of mouth, repeat customers and people who searched your name after seeing your sign do not show up as anything useful. Read the numbers as a partial picture, because they are one.
And the changes themselves stay with you. The AI can draft a new page title for you to review, but you check it against what the page actually offers, and you are the one who publishes it. Note the date you made each change, so next month's read can tell you whether it worked.
An editorial example
This is an invented example to show the read, not a real business.
Say you run a two-person bicycle repair shop in a mid-sized town. A good visit for you is a booking through the online form or a phone call. You export the three reports for the last 28 days against the 28 days before and paste the brief, noting that you added a page about e-bike repair five weeks ago.
The report comes back with the queries that changed most, and two of them are e-bike searches that did not appear in the earlier period. That matches the new page, so the change is real and explained. Under the near misses, it lists a query for flat tire repair in your town, where your page shows often, sits around position six, and earns very few clicks. It notes the page title says only "Services" and suggests a title that names flat tire repair and the town.
You open the Search Console file, find the row, and the numbers match. You check the page, and the title is in fact "Services." That one is worth doing today. The second action, writing more blog posts about bike maintenance, names no page and points to no row, so you skip it.
The whole read took under an hour, and you left with one change you trust and a date to check it.
Which export answers which question
| The question you have | The export to read |
|---|---|
| What do people search for when they find me? | Search Console, Queries tab |
| Which of my pages does Google show most? | Search Console, Pages tab |
| Where are people being shown my site but not clicking? | Search Console, Queries tab, sorted by impressions |
| Which pages do visitors land on, and do they book or call? | Google Analytics, landing page report |
| How much of my traffic comes from search versus social or direct? | Google Analytics, traffic acquisition report (an optional fourth file) |
If you only have time for one, start with the Search Console Queries export. It is the one most likely to hand you a change you can make the same day.
When to read it again
Run the same brief next month with fresh exports, and add one line listing the changes you made and the dates you made them. Ask the AI whether those pages moved. A page title change usually needs a few weeks before Search Console shows a difference, so give it the full 28 days before you judge it. Keep the brief in a note so the second read takes ten minutes to set up.
If search is the channel you are working on, our guide to AI search optimization for local businesses covers how the same pages show up when people ask an AI for a recommendation. If you want a starting list of the searches your buyers use, the buyer intent keyword generator gives you one to compare against your Queries export.
Of every job on this list, social media is the one that comes back every week, and Boomp is software that does that one for you. See ten posts made from your website, free
