Be the answer, not the tenth link.
When someone asks an AI assistant who to hire, it replies with a paragraph — often with no links at all. There is no page two. Either the model knows your business well enough to name it, or you are invisible. GEO is the work of making sure it knows.
SEO ranks a link. GEO becomes the sentence.
The habits that win at search do not automatically win here, because the reader changed. A crawler takes your page, a model compresses it, and a human never sees the original.
| SEO | GEO | |
|---|---|---|
| Unit of success | A ranked URL | A sentence inside a generated answer |
| Who reads the page | A human, after clicking | A crawler, then a model, then maybe a human |
| What wins | Links, keywords, page speed | Extractable facts, structured data, an unambiguous identity |
| Format that works | Long pages, keyword coverage | One question, one direct answer, machine-readable markup |
| Failure mode | Ranked eleventh | The model confidently states something wrong about you |
That last row is the one people underestimate.
In search, being absent is the bad outcome. In AI answers, being described wrongly is worse — a model that tells people your old opening hours will keep doing it until the sources it learned from change.
Three layers, each one earning the next.
Skipping a layer does not speed things up — it just makes the next one produce numbers you cannot trust.
Make what you already have machine-readable
Crawlers cannot guess. Structured data, clean headings, a reachable robots.txt and a published llms.txt decide whether a model can extract a fact about you at all.
Publish content built to be quoted
One page, one question, one direct answer — served as HTML and as plain Markdown, indexed in llms.txt. Set to allow AI crawlers while staying out of Google, so it never competes with your own site.
Measure what AI systems actually did
Server-side logs of real bot requests, real referral visits and real attributed leads. Not an estimate, not a sample — the requests that actually hit your pages.
Five stages, and the order is the point.
This is not the order the tabs appear in — it follows the dependency chain, because each stage produces what the next one needs.
Diagnose
GEO AuditCrawl your site, score five dimensions, and turn the findings into draft assets you can import with one click.
Fix the foundations
Readiness AnalysisThirteen pass or fail checks across technical basics, business information and AI readability. Re-run after each batch of fixes.
Publish for extraction
AI Content NodesTurn the audit questions into published question-and-answer pages, reviewed before they go live and rebuilt automatically when your facts change.
Measure
AI Crawler TrafficWhich bots came, which pages they took, which pages they ignored, and how many visitors arrived from an AI answer.
Monitor
GEO AnalyticsWatch the 30 and 90 day trend, keep published facts accurate, and push events into your own systems when something changes.
Tool links open the workspace and require a signed-in account.
Every tool, and the job it does.
GEO Audit
Enter a URL, get a five-dimension score with the evidence behind it, plus importable draft assets.
Schema 25 · Content 25 · AI Visibility 20 · Meta 15 · Citation 15
Readiness Analysis
Thirteen equally weighted checks on one page. Four of them read your Brand Voice Hub, so curating verified facts moves the score.
Good 67+ · Developing 40–66 · Urgent under 40
AI Content Nodes
Question-and-answer pages in eight formats, with a review workflow and automatic rebuilds when the underlying facts change.
FAQ · Service · Product · Spec · Policy · How-to · Comparison · About
AI Crawler Traffic
Thirty days of real bot requests: coverage, orphan pages nothing has fetched, vendor mix, crawl intent and response health.
Server-side logs, not estimates
GEO Analytics
Seven, thirty or ninety days across crawls, citations, AI-sourced visits and attributed leads, filterable by platform and surface.
Five charts · ROI funnel · Engine breakdown
Fact Check
Compare the facts you have verified and published against what public sources say about you, and turn any mismatch into a correction task.
Only verified, publicly shareable facts are checked
Tracking Tasks
Run a set of questions on a schedule instead of by hand, with a run ledger and webhooks into your own systems.
Hourly · Daily · Weekly
Webhooks
Five signed events, so a citation or a first crawl can trigger something on your side without anyone watching a dashboard.
HMAC-SHA256 signed · Replayable
Which signal to trust first.
AI crawler traffic
Every row is a request that actually reached your server, with the agent that made it and how long you took to answer. Start here — it is the one number nothing else has to interpret for you.
The most useful view is the orphan list: pages you published that no bot has ever fetched. A list that will not shrink is a discovery problem, not a writing problem.
Rates versus trends
The headline percentages on the hub are lifetime averages, so they get harder to move as your history grows. To judge recent work, read the 30-day trend in GEO Analytics instead.
A dash is not a zero. It means not measured; zero means measured and genuinely nothing. That distinction is kept everywhere in the product.
What to do, and how often.
Most of it is not a daily job. The parts that repay attention are the weekly coverage check and re-auditing after anything changes on your site.
| When | What |
|---|---|
| Once, at setup | Complete the business profile, verify brand facts, publish llms.txt, run the first audit in discovery mode, import the node blueprints. |
| Weekly | Check whether the orphan-node list is shrinking, approve pending nodes, clear fact mismatches. |
| Monthly | Re-run Readiness on key pages, read the 30-day trend and the ROI funnel. |
| Quarterly | Re-run the full audit and read the comparison panel against last time. |
| After any site change | Re-audit the homepage and confirm robots.txt still allows AI crawlers — the highest-impact regression there is. |
What each action costs
Credits are shown next to the step that spends them, so the bill is never the first place you learn about it.
| Action | Credits | Notes |
|---|---|---|
| GEO Audit | Free | Crawling, scoring, assets and exports are unmetered. |
| Readiness analysis | 1 | No AI call — the cheapest way to confirm a fix landed. |
| AI node generation | 2 | Nightly rebuilds of stale nodes are free. |
| Node citation test | 1 | One question, one engine. |
| Fact check | 1 | Charged per verified fact, so the count is shown before you run it. |
| Tracking run | 2 | Per question, per engine, per run — size the set deliberately. |
The one that multiplies
A tracking run charges per question, per engine, every time it fires. Ten questions on two engines running daily is six hundred credits a month. Start with the handful of questions you would actually act on.
Start where the evidence is.
Run one audit on your homepage. It costs nothing, takes a minute, and hands you the list of questions worth publishing answers to.