Search analytics is a list of what people wanted from you

Search analytics is a phrase with two working meanings. The first is search engine analytics, sometimes called Google search analytics: the record of which queries put your pages in front of people on a search engine. The second is site search analytics: the record of what visitors type into the search box on your own site, app or store. A third phrase, search-driven analytics, names a business intelligence product category and belongs to a different conversation.

Search Console is the usual source for the first sense. It lists, per query and per page, impressions, clicks and average position. The report has widened. Ed Latimore reported that Search Console now shows AI search results as well, and that Latimore's site drew 95k impressions from those in 28 days.

Site search is the second sense, and it lives in your own tools. The data is what visitors typed into the box, including the searches that returned nothing. Sebastian Caniulao, who works in ecommerce email and growth, noted that in Shopify that report sits under Analytics as "top online store searches with no results".

Both senses produce the same object: a list of things people wanted from you. The useful part of that list is the rows where you had nothing to give them.

The first number on the Search Console screen is impressions, and it inflates. Aritra, of the company Supaboard, reported their Search Console reading 5.4M impressions and about 10 clicks a day. Aritra explained the gap as "Most of that surface is people getting the answer and never arriving", and the metric itself as "The channel measures the leak, not the flow."

Smaller sites show the same shape. One builder documenting a site day by day reported 101K impressions, a 0.3% CTR and a 24.7 average position, against 270 clicks on the day. The builder's own verdict on the headline figure: '101K sounds better without the word "impressions."'

The AI rows in Search Console leave this intact. Ed Latimore reported that the top-cited pages on Latimore's site in AI results "strongly correlate with pages that are also ranked in classical SEO", and drew the conclusion that "There is no such thing as specific AEO or GEO." On that site, the page that earned a citation was the page that already ranked. A new label on the column changed nothing about what it counted.

The reverse case. Yuri Queiroz built a local clinic's site and reported the last 90 days as 153 clicks, 12K impressions, 1.3% CTR and 7.4 average position. Queiroz was asking for advice on the CTR, and reported the client was very happy, because "new customers are finding the clinic through Google almost every day". The 153 is the business number. Clicks, and what happened after them, are the only rows in the report that touch revenue.

The query column is the useful part. One founder argued that Search Console gets read as an SEO dashboard while "almost nobody treats it as a demand sensor". Because it "only shows queries that already hit your own pages", it "can tell you what your audience wants that you haven't built". Three rows: accidental rankings at position 15 and beyond with impressions and zero clicks; queries containing "tool", "generator", "template" or "alternative"; and queries now answered inside AI overviews.

Elchanan Uzan, an app developer, reported 70 impressions in 7 days, 0 clicks and average position 60, every query the exact question the app answers: "What is system data on mac". Uzan's reading: "The demand is real and I am not in front of it." Brooks Conkle opened Search Console on a small site and saw people across Alabama searching for a CPA: "So I changed one thing. 2 leads so far."

Larger sites export. Tim Bennetto, who builds the tool Pallyy, reported organic impressions up from 41K to 140K a day, partly from "exporting queries from GSC to find ones we rank for but dont have pages, then build those pages". Rashid Khasanov, who runs angelmatch.io, exported a CSV from Search Console and updated content, metadata, FAQs and internal links on the pages "getting the most impressions": "100 pages were updated in 1 week."

The list locates demand and cannot create it. Deepak Yadav reported 124 pages, 130 indexed, 6 driving 94% of clicks, all hand-written with real data. The 118 AI-built pages rank. "Nobody searches." Yadav's line: "AI scales content. It does not scale demand."

The same list exists on your own site. John Affourtit, at the payments company Flex, claimed that before Flex launched HSA and FSA checkout with the retailer Dermstore, "HSA" and "FSA" were already among the retailer's top on-site search terms: "The demand was there". A few months in, by the partner's own figures rather than the retailer's, those orders ran 43% higher in value than the site average, 30% of the customers were new to the retailer, and orders carried 24% more items.

Smaller operators read the box too. The team behind the directory Prop Firm Map reported that '"Is [prop firm] legit?" shows up in our search data every day', and built a Trust Check page for every firm. The UK firearms marketplace Rightgun wrote that "Our search data shows steady UK interest in the Blaser R93", and published a buyer's guide.

Zero-results searches are the sharper half. Sebastian Caniulao, commenting on someone else's bookstore, one planning to import 14 million titles, called the zero-results report "a demand list written by customers, which beats any bestseller chart as an ordering signal", added that "almost nobody reads it", and proposed importing titles on demand off that report. A zero-results search is a customer telling you what to stock, or what to call it, in their own words and for free. Another founder reported that "something to clean my garden" returned 0 results on a shop that sells exactly that, because shop search is "mostly string matching on titles and tags".

The sensor records noise as well. Chris M. Walker, of the freelance marketplace Legiit, reviewed months of internal searches and found the kind outnumbering every other was "freelancers searching for their own services or profile." Strip the self-searches before acting on the rest.

Aritra, of Supaboard, also argued that "measurement is trustworthy on direction and not on level". Aritra reported that Search Console and GA4 disagree on the same sessions every month, put it down to sampling and timezone boundaries, and gave the verdict "i will bet on the direction".

The level moves on its own. Nate Tucker reported that Search Console tracks the http and https versions of a site separately: the http version ranked around position 9, the https version around 27, "That's with a redirect, too." Sahil Panhotra, an indie builder, found 37 pages not indexed and 1 indexed, because every page named two addresses for itself, with and without a trailing slash, and one redirected to the other. Panhotra called it "the scariest kind of bug, it looks like nothing's wrong because for you, nothing is", and fixed the canonical tags. The recipe site Inspired Taste reported a September dip in Search Console clicks absent from ad revenue, Google Analytics and Cloudflare referrer reports; Glenn Gabe replied with no such dip across sites. A developer building a reporting tool claimed Search Console deletes history after 16 months.

Site search has its own failures. Jan Amann, who maintains the next-intl docs, reported that on the docs' Algolia search "the amount of search queries keep dropping", because developers now ask an AI assistant instead. Back in 2023, Taylor Otwell, who created Laravel, found a docs page with 1,008 headings unindexed, a cap Otwell put at 1,000 headings per page. A falling row, or a missing one, can be the tool's doing.

Nick Skomor, who does marketing at the company EchoTrade, described a setup built around an AI assistant. The headline figure is impressions, about 9x in 3 months by Skomor's own claim, and impressions were discounted above; the part worth copying is the queue the setup builds. The assistant was connected to Search Console, GA4 and Ahrefs. Each day it reports position moves, new queries and indexing issues, a pulse Skomor put at thirty seconds to read. It pulls "queries sitting at position 5-15 that already have impressions", and Skomor traced the early wins there: "This is where our first rankings came from." Drafts "come from GSC gaps instead of guesses", and a Monday report sets three priorities. The caveat came from Skomor as well: the link recommendations still get checked by hand every time, because "It's confidently wrong there more than anywhere else."

The site search side runs on the same rhythm. CrimeLine, a UK legal publisher, shipped a refinement to its search box after readers had found the old one frustrating. Put together, the accounts above make a loop of five items, run once a week:

1. Export the queries that have impressions and no page of their own.
2. Read the zero-results list, and strip the self-searches before anything else.
3. Take the queries sitting at position 5 to 15.
4. Ship one page or one small tool, and add the products people typed.
5. Check any level against a second source before it goes in a report.

The loop is dull, and it is weekly, which is why it gets skipped. It is also the kind of dull weekly work that can be handed to an agent, with the link recommendations still checked by hand.