Ask your team a simple question: when a buyer asks an AI assistant for the best options in our category, are we in the answer? In most marketing teams, nobody knows. The SEO report covers rankings, clicks and backlinks. It says nothing about the place where many shortlists now start.
Generative engine optimisation, or GEO, is the name that stuck for the work of getting your company named and cited in those answers. Most of what is written about it is either vague ("create great content") or speculative. This piece is the method we use, step by step, so you can run it with or without us.
Why B2B teams should care now
A B2B buyer rarely types a single keyword and clicks the first result. They ask a question with context: "What are the alternatives to our current agency for a lean marketing team?" or "Which tools help a founder run paid search without hiring?" Assistants answer that kind of question with a short list, a few sentences on each option, and often a set of cited sources.
Two things follow. First, the list is short, so being absent is close to being invisible for that question. Second, the answer is assembled from pages the assistant could find and understand. That second point is the good news, because it means most of the work is on surfaces you control.
Step one: build the question set your buyers actually ask
Start with questions, not keywords. Pull them from the places buyers already tell you what they want to know:
- Sales call notes and demo recordings. The questions prospects ask in the first ten minutes are the questions they asked an assistant the night before.
- Lost-deal reasons. "We went with X because..." tells you which comparison questions you are losing.
- Search Console queries. Long, question-shaped queries are a direct read on how people phrase the problem.
- Support and onboarding tickets. These surface the "how does it work" questions that shape consideration.
Group them into three buckets: category questions ("best options for..."), comparison questions ("X vs Y", "alternatives to X") and problem questions ("how do I fix..."). Aim for a set you can realistically re-run on a schedule. A few dozen well-chosen questions beat several hundred you never look at again.
Step two: sample the answers, and sample them more than once
Run each question through the assistants your buyers use. Record three things per answer: whether you are named, whether a page of yours is cited as a source, and which competitors and third-party sites appear instead.
Run each question more than once and on more than one day. Assistant answers vary between runs, so a single screenshot proves very little. What you want is a trend: over repeated samples, how often are you present for this question, and is that rising or falling? Treat it the way you would treat a rank tracker, not a one-off audit.
Step three: read why you are missing
When you are absent, the cited sources usually tell you why. The common patterns:
- No page answers the question directly. You have a product page and a homepage, but nothing that plainly says who you are for, what you replace and how you compare. Assistants cite pages that state the answer, not pages that imply it.
- The answer exists but is buried. The key sentence sits in the fourth paragraph under a clever headline. Put the direct answer near the top, in plain words, then support it.
- Third parties define your category and you are not in their lists. Review sites, directories, comparison articles and community threads show up as sources again and again. If you are not on them, you are relying on your own site alone.
- The page cannot be read. Content rendered only after heavy client-side scripts, pages blocked in robots.txt, or key facts locked in images all make a page harder to use as a source. Check that you are not blocking the crawlers the assistants rely on unless you mean to.
- Your facts are inconsistent. Different pricing on two pages, three descriptions of what you do, an old product name still live. Inconsistency gives an assistant no clean statement to repeat.
Step four: fix in the order that pays
Resist the urge to publish twenty new articles. The order that usually pays back fastest:
- Make your core pages answerable. Homepage, product, pricing and the one page that explains who you are for. Each should state, in the first screen, what you do, for whom, and what you replace.
- Publish honest comparison and alternatives pages. If buyers ask "alternatives to X", a clear, fair page that names the trade-offs is exactly the kind of source that gets cited. Unfair comparisons tend to backfire with readers anyway.
- Add structure where it fits. Clear headings phrased as the question, short direct answers under them, and structured data where the page type supports it.
- Close the third-party gaps. Get listed accurately on the directories and review sites that keep appearing as sources, and correct any that describe you wrongly.
- Then write new content for the problem questions where no good answer exists yet.
Re-sample after each batch of changes. The measure is presence over time for your question set, alongside the usual Search Console numbers, because the changes that make a page easy to cite usually make it clearer for search visitors too.
What not to do
Do not stuff pages with invented statistics in the hope an assistant repeats them. Do not create thin pages for every phrasing of a question. Do not buy placement on low-quality "best of" lists. All of these are cheap to do and expensive to undo, and none of them survives a buyer who checks.
How the Search Crew runs this
In Spryxa, this is the Search Crew's weekly job. Vega scans keyword movement and new GEO opportunities, and produces a search opportunity brief. When the evidence supports a concrete change, the brief carries a list of website fixes, each with a target URL, a rationale, the evidence behind it and a risk rating.
What happens next depends on risk. Low-risk site fixes, like tightening a title or adding a missing direct answer, can ship automatically inside guardrails you set. Anything riskier is staged for your review. New pages and rewritten copy go through the Content Crew, and a person reviews the draft, the claims and the publishing destination before anything goes live.
The crew is measured on Search Console clicks, impressions, CTR and average position, plus verified outcomes on the specific changes it made. That learning is strongest when Search Console and site-change permissions are connected, so start there. You can see what connects today on the integrations page, and the step-by-step versions of these jobs are in the how-to hub.
Where to start this week
Write down ten questions your last ten prospects asked. Run them through two assistants, twice each. Note where you appear and which sources show up instead. That one exercise will tell you more about your AI visibility than any dashboard, and it will tell you which page to fix first.
If you would rather have the crew run the first pass, start with the free audit. It reads your site the way a crawler does and shows you which pages are ready to be cited and which are not.