1,683
Fixes implemented across 8 sites
Service · SEO & GEO
SEO built for humans and machines.
SEO advice is cheap. Implementation is where it falls apart, because the fixes that matter are technical and most reports stop at telling you what is wrong. We do the other half, plus the newer job: getting your pages cited by AI search.
Fixes merged, not listed
Crawler access verified
Citation tracking you keep
Why a page does or does not get quoted
We check the gates in order. Fixing passage structure on a page the crawler cannot reach is wasted work.
20wks
Measured in our own benchmark
519
Hyperliquid Guide pages earning impressions
35,723
Clicks in 28 days on that one site
The gapWhat a normal audit gives you What we hand over
Everyone sells the audit. Nobody merges it.
The reason SEO retainers stall is rarely that the advice was wrong. It is that a list of 40 issues lands on a dev team who already have a roadmap, and it sits there until the contract renews. Across our own portfolio, twenty weeks of weekly measurements put 91.9 percent of generated recommendations into production. That number is only that high because the merging was automated.
- A PDF of issues ranked by a severity nobody agreed on
- Advice your developers still have to translate into changes
- Rank tracking that cannot see an AI citation at all
- No check on whether AI crawlers can even reach the page
- A retainer that renews whether or not anything shipped
- The fixes implemented and deployed, not described
- Crawler access verified per bot, with the response codes
- Key sections rewritten into passages a model can quote whole
- Schema, snippet caps, and llms.txt in place
- Citation and position tracking you keep after we leave
How it runs
A report you can actually merge, every week.
Week 0
Scope call
We start with whether the thing is worth building at all. If we are the wrong shop for it, we would rather say so in week one than in week nine.
Weeks 1–2
A thin slice, end to end
One real path all the way through the system. Ugly, but running and instrumented, so the unknowns surface while there is still time to do something about them.
Weeks 3–11
A demo you can poke at, not a status update
Something real every week. Nothing gets saved up for a reveal at the end, so a change of direction costs a conversation instead of a change order.
Final week
Handoff with runbooks
Code, deploy steps, and what to do when it breaks at 3am. We stay available for follow-on work, but we do not build in a dependency on us.
A page that ranks but never gets quoted is losing a race nobody told it about.Concept211 · Technical SEO and GEO
Photo by Paul Lichtblau on Unsplash
Run on
The platform we sell is the one we use.
ConceptSEO runs the weekly cycle on our own sites first. Client work goes through the same pipeline, not a spreadsheet.
LaravelConceptSEO
PHPAnalysis workers
OpenAIPassage rewriting
Claude SDKAudit reasoning
Proof · our own site
Read the case study
We ran this on ourselves before selling it.
Getting the Hyperliquid Guide quoted by AI search
An independent reference with 50,000-plus readers a month, turned into a page AI engines cite. Server-rendered passages, schema, verified crawler access, and the snippet caps lifted. The write-up names what was measured and with which tool, so you can check rather than take our word for it.
519Pages earning impressions
35,723Clicks, 28 days
614Fixes implemented
Asked before hiring
Three questions, and what we do about each.
These are the questions people are searching when they land here. The mechanism behind them is written up at length in the GEO methodology article. What follows is the other half: what this practice does about each one, and what you are holding at the end of it.
What content signals make a page more citation-ready?
Self-containment first. An engine lifts a chunk, not a page, so a paragraph that needs the three above it for context cannot be lifted at all. We rewrite the sections that matter into passages that lead with a direct answer under a plain-question heading, put a short standalone summary near the top, and attach a source or a stated method to every figure on the page. An unsourced number is the thing a model declines to repeat.
You get the rewritten copy in your own templates, stable ids on every heading and every headline figure so a third party can link to one claim, and a line on the page saying which claims are documented and which are first-hand. What changes is that the page becomes quotable in one piece rather than requiring a summary nobody will write.
What structured data helps AI search engines cite my pages?
The types that genuinely describe the page, wired together rather than scattered. In practice that is Article on editorial pages, Organization and Service on the pages that sell, BreadcrumbList on everything, and FAQPage only where the questions are visible on the page. The part most sites miss is the wiring: giving each node a stable id so the article, the breadcrumb and the organization reference one another instead of floating side by side, and pointing an article's subject at the entity it is about on that entity's own domain.
You get the markup in your templates, validated against the pages it describes, and generated from the same content the reader sees so the two cannot drift apart. What changes is that an engine stops inferring what the page is and what it claims.
Why does one page get mentioned in AI answers and a similar one not?
Nine times in ten it is access or shape, not quality, and the two are easy to tell apart once you look in the right order. Access: can each AI agent actually fetch the URL, and does the raw response carry the words. Shape: is there a passage on the page that answers the question by itself, and is a snippet cap throttling how much can be shown. We run those checks per agent with the response codes recorded, then read the edge logs by path and verified-bot status rather than by user agent, because the headline refusal rate on most sites is dominated by scanners borrowing crawler names.
Only after that does the answer become editorial, and the checklist for that part is the six levers in the FAQ below. You get the per-agent response codes, the log read, and citation tracking that keeps reporting which of your pages get quoted, for which prompts, on which engines, after we leave.
Measured · with the window on it
What this actually produced, with the working attached.
Five numbers, each read off a query you can re-run rather than a claim you have to take. The window sits next to every one of them, and the link goes to the page that carries the full method. Where a figure moved the wrong way it is printed the way it came back.
| What was measured | Result | Window and source |
|---|---|---|
| This site in Search Console | 518 impressions across 19 URLs, and 0 clicks. The preceding 28 days held 64 impressions. | 26 Jul to 22 Aug 2026 against 28 Jun to 25 Jul, read from the Search Console Search Analytics API on 25 August 2026 |
| Recommendations implemented across eight sites | 1,683 closed, and 1,642 of them shipped end to end by the agent rather than by a person editing code | 20 weeks to 28 July 2026, from the recommendation ledger. Published in the benchmark |
| Where that work landed | 41.0% in content and internal linking. Schema, images and PageSpeed together came to 26.9%. | Same ledger, n = 1,683, broken out by source analysis |
| AI-crawler access, fetched per agent | 32 of 32 homepage fetches as GPTBot, ClaudeBot, PerplexityBot and Google-Extended returned HTTP 200, with zero variance | Eight sites, 28 July 2026, reproducible with curl. See what did not vary |
| What a raw refusal rate is counting | 74,236 crawler requests read at the edge across 14 properties. On this site, 1,171 of 1,180 requests carrying the Applebot string asked for files it has never published. | 4 to 10 Aug 2026, Cloudflare edge analytics, read by path |
The zero is not a typo and it is not buried. This site earns impressions and no clicks, on queries where it sits at position 35.8 for the term this page targets, so the honest read is that the work has moved visibility and has not yet moved a single visit. The earlier 28-day window is derived by subtracting the 28-day totals from the 56-day totals rather than requested directly, which makes it the weaker of the two numbers. Search Console filters page-level rows, so every total here is a floor. The eight-site figures are ours and our clients' rather than a random sample, there is no control group behind them, and the sites are anonymized in the write-up.
Before you ask
Questions we get every time.
The things people ask on the first call, answered before you have to ask them.
Ask us something elseWhat is GEO (Generative Engine Optimization)?
GEO is the practice of getting a page quoted as a source by AI search engines rather than only ranked in the classic ten blue links. When ChatGPT, Perplexity, or Google's AI Overviews answer a question, they read a handful of pages, write one answer, and cite the sources they leaned on. GEO is the work of making your page one of those cited sources: readable in the raw HTML, structured into liftable passages, and marked up so a model can trust it.
How is GEO different from traditional SEO?
Traditional SEO fights for position in a ranked list a person scans and clicks. GEO fights to be the passage a model lifts into a written answer the reader often never clicks past. The two are not rivals. AI engines pull heavily from pages that already rank on page one, so classic SEO sets the ceiling and GEO sits on top of it. We do both: the crawlability, indexation, Core Web Vitals, schema, and site structure that move rankings, then the passage work that wins the citation. The long version, with what actually changes in the day-to-day work, is in what changes between GEO and traditional SEO.
How do you get a page cited in AI Overviews, ChatGPT, and Perplexity?
Six levers, run in order. Confirm the AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended) get a clean 200 instead of a firewall block. Serve the real content in the raw HTML so a crawler that fetches the URL sees words rather than an empty shell. Rewrite key sections into self-contained passages a model can quote whole. Then schema, an unlimited max-snippet cap, and an llms.txt. The full checklist is in our GEO methodology article.
How do you track AI-search citations?
AI citations do not show up in a normal rank tracker, so we run citation tracking with ConceptSEO, the platform we built for exactly this. It reports which of your pages get cited, for which prompts, on which engines, and how that moves week to week, next to classic positions. Watching position and citation share together is how you tell a page that ranks but never gets quoted from one that has started showing up as a source.
Do you implement the fixes or just report them?
We implement them. An audit that lists 40 issues nobody has time to merge is the standard failure of this industry, and it is why most SEO retainers quietly stall. You get the changes shipped, plus the tracking to see whether they worked.
Work with us
Start a project Ranking, but never getting cited?
Both halves are solvable: the technical fixes merged rather than listed, and the passage work that wins the citation. You keep the tracking either way.