The search and AI analytics agent counts people, search crawlers and AI readers on every page, without cookies or stored IP addresses. It combines daily totals with page audits, connected search data and AI citation checks, then writes a fix brief. It never edits your site.
Inside the Search & AI Analytics Agent
Follow requests from the server to daily totals, then to the evidence a developer can act on.
Sort page requests into six reader classes using the user agent each request declares. That identity is not verified.
Daily totals meet nightly audits and connected search, speed and citation sources. A missing source stays “not connected”.
The output names the weak page, the evidence and how to verify a fix. This agent never edits the page.
Daily page totals, with no stored IP address or person identifier. Classification is based on declared user agents.
Google Analytics runs in the visitor’s browser, and in Europe only after they accept cookies. So it misses everyone who declined — and it misses every crawler, because crawlers do not run its script. For most of the last twenty years that did not matter much. Now a growing share of the reading that decides whether you are found is done by machines: AI crawlers building the indexes AI search answers from, and AI assistants fetching your page in the moment a person asks them a question.
None of that appears in a browser-based report. You can rank, be crawled, be read by ChatGPT a hundred times a week and be cited in its answers, and your dashboard will show a quiet Tuesday. Or the opposite: a firewall setting quietly blocks every AI crawler, and nothing anywhere tells you.
This agent counts on the server, where every request arrives, and keeps the count as plain daily totals. Then it does the thing nobody has time for: it puts that next to Search Console, page speed, an audit of every page and a weekly check of who the AI engines cite, and says which pages need attention and why.
every request, no cookie
people · search · AI
nightly · every page
Search Console · speed · GA4
weekly AI citation checks
score · digest · fix brief
Each page view is counted on the server as it is served, into daily totals by page, visitor class, crawler and source. No IP address is stored, no identifier is created, and no cookie is set to do it — so it counts the visitors who decline cookies and every crawler that never runs JavaScript, which is all of them.
Every request is sorted into one of six classes: a person, a search engine crawler, an AI training crawler, an AI search crawler, an AI assistant fetching the page because someone asked it a question, or another bot. That last-but-one class is the one to watch: it is a model reading your page to answer a person in real time.
A nightly crawl checks each registered page the way a search engine and a language model would read it: title and description, one heading, canonical, structured data, an answer near the top, internal links, the llms.txt entry. Each page gets a search score and an LLM-readiness score, with every failing check named.
Where they are connected, it pulls Search Console daily — web, image, video, news and Discover, with up to sixteen months of history — PageSpeed with real-user Core Web Vitals, and Google Analytics engagement for the visitors who consented. A source that is not connected says so, rather than showing a zero.
Once a week it puts a fixed set of English buyer questions to the connected AI search engines and records which pages each answer cites, alongside the page that should have been cited. That is the closest thing there is to a ranking report for AI answers.
Everything folds into one 0–100 health score per page, weighted across search, LLM readiness, speed, traffic and AI attention, with the weight of a missing source redistributed rather than scored as zero. A Monday digest lands by email, and every weak page has a brief a coding agent can act on: each failing check, where in the code the fix goes, and how to verify it.
Every request is sorted by the user agent it declares. That is the honest limit of any server-side count: a crawler that lies about who it is will be counted as whatever it claims. The major AI and search crawlers announce themselves, which is what makes the split useful.
The point of joining the sources is that each one answers a different question about the same page, and the answer is usually in the gap between them.
Was it crawled? Search engine and AI crawler fetches, per day. No recorded AI fetches in a week is a reason to check crawl access. It does not prove a page is absent from AI answers: an engine may use an earlier index or another retrieval source.
Was it read by an AI for someone? Assistant fetches are the nearest thing there is to seeing an AI answer being written from your page.
Was it cited? The weekly probes show which page an engine cited for each buyer question, and whether it was yours.
Did it show, and did anyone click? Search Console impressions, clicks, position and the queries behind them, daily, with up to sixteen months of history.
Did a person read it, and then get in touch? Human views and entrances by source, and — only when someone sends an enquiry — the page they landed on.
The count is a set of daily totals: this page, this day, this class of visitor, this source, so many views. There is no row for a person to be found in, because no IP address is kept, no identifier is made and no cookie is set to do the counting. Signed-in staff, link prefetches and our own audit crawler are left out, so the numbers are about the outside world.
That is also why it can count everyone. A tool that needs to recognise a returning visitor needs permission to do so. A tool that only adds one to a daily total does not need to recognise anyone. Our own privacy policy and cookie policy describe exactly this, and they are the documents to check it against.
Every weak page gets a brief written for a coding agent or a developer: each failing check in order of importance, the fix, where in the code it belongs, the search queries the page nearly ranks for, the AI questions it is not cited for, and the command that proves the fix worked. A site-wide problem gets one brief across every page it affects; a problem in the CDN or firewall is marked for a person instead.
On our own site the same evidence feeds the SEO and AI visibility agent, which drafts changes for an editor to approve and then measures them. Kept apart on purpose: the agent that counts should never be the one that benefits from the count going up.
The analytics agent never changes a page: it counts, audits, scores and writes the brief, and a person, or a separate agent a person approves, decides what to do with it. It cannot verify that a crawler is who it says it is, because visitor classes come from the user agent as claimed. It does not replace Google Analytics for the visitors who consent to it; it adds everything a cookie-based tool cannot see.
Every page in this portfolio carries one of these. The claim a page refuses to make is the reason to believe the ones it does — see our work.
Every source is optional. One that is not connected renders as not connected, never as a zero.
Google Analytics runs in the visitor’s browser and, in the EU, only after they accept cookies, so it misses everyone who declines and every crawler, because crawlers do not run its script. This agent counts on the server as each page is served, with no cookie, so it sees people who declined, search engine crawlers, AI training crawlers and AI assistants reading a page for someone. It keeps Google Analytics for what it does well — engagement among consented visitors — and puts both next to Search Console and the AI citation checks.
Yes. Every request is sorted by its user agent into six classes, including AI assistants fetching a page because a person asked a question, AI search crawlers and AI training crawlers, and the counts are kept per page per day. The class comes from the user agent as the crawler declares it; it is not verified, so a bot that lies about who it is will be counted as whatever it claims.
The counting itself sets no cookie and stores no identifier or IP address — only daily totals — so it does not wait for consent. On our own site the only thing that does is Google Analytics, which loads only after a visitor allows it. Whether the same holds for your site depends on what else it runs, so your counsel should confirm it. This is engineering guidance, not legal advice.
One number from 0 to 100 per page, combining the search audit, LLM readiness, Search Console performance, speed, human traffic and AI attention. Each component has a weight; if a source is not connected or has no data for that page, its weight is shared among the others rather than counted as zero, and every score records the weights it used.
Usually because something in front of the site is turning them away. Many CDN and firewall settings now block AI crawlers by default, which overrides robots.txt, and bot challenges can stop them too. The agent flags a page with no AI fetches in seven days so you can check whether you are blocking the very engines you want to be cited by.
No. It is read-only by design. It writes a brief for each weak page — every failing check, the fix, where the fix goes in the code and how to verify it — that you can hand to a developer or a coding agent. On our own site the SEO and AI visibility agent picks up the same evidence and drafts changes for a person to approve.
Yes, as a managed engagement. The counting, the classes, the nightly audit and the Search Console and PageSpeed connections carry across to most sites; the setup is connecting your sources and checking your CDN is not blocking the crawlers you want to count.
What we built, what stage it is at, and a link to the live thing wherever there is one.
yourOS is an operating system with an AI at its heart. You talk to it like a person, out loud or typed. It makes your calls, plans your days and works on your projects with a crew of agents, on your own private server. Our own product, built from the first commit; the waitlist is open.
Website, backend and app assets for Sonora, a free AI sound-therapy app on iOS and Android published by Cognitive AI.
Brand, website, prototype, apps on five platforms, the AI backend and the algorithm-intelligence pipeline for Podswap, a social growth engine that runs without a human in the loop.
In closed beta: Roxy, a private AI assistant with permanent memory running on infrastructure the user owns. Brand, site, product, backend and conversational design.
Live at coherence.com: brand identity generated with AI agents we directed, the motion system and website for the group behind Sonora, AboutAI, WakeUp and MudShadows. Agent work remains in build.
Live at domainbooth.com for DomainBooth, a premium domain brokerage: a new animated mark, every page redesigned and rebuilt with designed mobile layouts, and a custom CMS for its sales record, featured names, press and enquiries.
Tell us about your site and what you already measure. We will tell you what a server-side count would add, and whether your CDN is turning away the AI crawlers you want to be read by.
30-minute call · no deck · we will say if the answer is no