Recently I presented Mastering AI Search for Startups at the NatWest Accelerator Hub — a session built around the framework in my book, covering why most startups are invisible in ChatGPT, Perplexity, and other LLMs, and what to do about it.
This post turns those slides into a practical guide you can apply immediately.
The problem is not what most founders assume. Your startup is not missing from AI-generated answers because you lack content, or because your team is not good enough. You are missing because AI systems cannot reliably read, understand, or verify what you have built. That is a structural problem. And it is fixable.
Check the slides for my presentation: Mastering AI Search Optimisation for Startups
Link to slides: https://docs.google.com/presentation/d/1yA7JdD8Z6v5GX84Dn8G1MMBJ9c0aDh7_aOGnXjY0tsE/edit?usp=sharing
What Is Generative Engine Optimisation?
Generative engine optimisation (GEO) is the practice of making your brand visible inside AI-generated answers — the short, definitive responses that ChatGPT, Claude, Perplexity, and similar tools produce when someone asks a high-intent question.
The old model was SEO: search, click, compare, decide. The new model is GEO: ask, receive an AI shortlist, decide.
That shortlist is everything. There is no page two. There is no “reasonably visible.” If your startup is not in the AI-generated answer, you are excluded from the decision entirely.
The data makes this urgent. According to a 13-month Search Engine Land analysis published in February 2026, LLM referrals convert at around 18% — higher than SEO, PPC, and email combined. The volume is still roughly one twenty-fifth of organic search traffic, but the conversion rate gap is closing fast. Getting into those answers now, before this becomes as competitive as traditional search, is one of the highest-leverage moves available to any startup marketing team.
Why Startups Get Passed Over by AI Systems
Before recommending any brand, an LLM needs to complete four cognitive tasks: categorise you, explain what you do, compare you to alternatives, and verify that you can be trusted.
Most startups fail at step one. The reasons are predictable.
The site is built on React, Next.js, or a no-code tool like Lovable or Replit. LLMs cannot execute JavaScript. Disable it on most startup homepages and you get a blank page. That is all the AI ever sees.
Every page shares the same title tag: just the brand name, nothing more. No topic, no signal, no context for crawlers. The blog is on Substack. The docs are on a subdomain. The app is somewhere else entirely. The brand is fragmented with no coherent map.
And there are no product pages. No canonical definition of what the product does, who it is for, and critically, what it cannot do.
These are not advanced problems. Most Web2 companies solved them years ago. Startups have not caught up.
The Four-Layer GEO Framework
Generative engine optimisation is built in layers, not tactics. Each layer depends on the one before it. Skipping ahead wastes the investment in everything that follows.
Layer 1: Technical — Can Machines Read You?
This is the foundation. Three questions matter here:
Can machines crawl you? Add a sitemap XML, a robots.txt, and an llms.txt file. Make sure nothing is blocking crawlers.
Is your content readable? Your core product content needs to exist as plain HTML. JavaScript-rendered pages are invisible to LLMs (like vibecoded apps with Lovable and Replit, but here’s a prompt to fix that).
If you are not sure, install the free AI Eyes or Detailed SEO Chrome extensions, disable JavaScript, and see what remains. If the page goes blank, LLMs are reading nothing.
Can AI parse your content? Add schema markup so AI crawlers can read your product in structured, table-like format rather than having to interpret narrative copy.
Six technical rules apply here: crawlable HTML, explicit crawl directives, one consolidated domain, one page per core concept, descriptive title tags on every page, and structured data throughout.
Until this layer works, nothing else matters.
Is your website invisible to AI? Run a free check with my Technical AI Visibility Checker tool.
Layer 2: Content — Can Machines Understand You?
Most startups publish announcements, changelogs, and abstract thought leadership. LLMs need something different.
AI systems are looking for definitions, mechanics, constraints, comparisons, and risks. Marketing narratives alone are not enough. “Our platform is secure and easy to use” gives an LLM nothing to work with. “Audited by [firm], SOC2 certified, with a 99.9% uptime SLA” gives it something citable.
The six essential content types for GEO are:
- Canonical explainers — “What is [your product]?” — clear, quotable, positioned at the top of every relevant page
- Spokes — individual pages covering risks, mechanics, and constraints, one concept per page
- Reference pages — fees, limits, technical parameters
- Product pages — what you do, who it is for, and what it cannot do
- Comparison pages — “X vs Y for [use case]”
- Negative qualification — “Who should not use this” — acknowledging limitations builds trust
When writing for LLMs, start every key page with a 150-250 word direct answer block. Use headings, bullets, and tables — these are the formats AI systems extract most reliably. Include original data wherever possible. Add FAQs with schema markup. Refresh core pages every 90 days to stay cited around 40% more frequently.
Fireblocks is a strong example of this done well. Their stablecoin content covers all six types: a glossary page, expert blog commentary, a State of Stablecoins research report, a product page, and a comparison table. That is why they dominate that category in AI answers.
Layer 3: Authority — Should Machines Trust You?
Trust is not what you say about yourself. Trust is what the rest of the internet says about you, consistently and independently.
What does not build AI trust: press releases about funding rounds, follower counts, logo walls.
What does: independent third-party mentions, appearing in educational content rather than just announcement coverage, and a consistent brand description across every source.
The formula for your canonical definition:
[Brand] is a [category] that helps [user] achieve [outcome] using [mechanism].
Write it down. Use it in every interview, every article, every podcast, every social profile. Do not vary it. LLMs build confidence by seeing the same clear explanation repeated across multiple independent sources. Inconsistency creates hesitation. Hesitation means you do not get recommended.
Named authors matter too. Anonymity is an authority killer. Every article on your site should have a real person attached to it — someone with credentials, a profile, and a track record.
PR is useful here, but not in the traditional sense. A press release about your Series B does less for AI trust than a piece of original industry research. A “State of [your market]” report, with named data and clear methodology, outlasts funding announcements by years in terms of how LLMs perceive your expertise.
Layer 4: Reinforcement — Is Your Explanation Repeated Elsewhere?
LLMs do not trust a single source. They trust patterns. One clear explanation repeated ten times beats one hundred pieces of scattered content.
The goal at this layer is to get the rest of the internet to say the same thing you say — consistently, independently, and across multiple platforms.
Not all channels are weighted equally. Based on a Semrush study of 230,000 prompts from October 2025, the top cited domains in LLM responses are:
- Wikipedia
- Medium
- YouTube
X (Twitter) sits near the bottom of the list. If your startup is spending its primary content budget on X, the return in AI search terms is close to zero.
Reddit is the most important channel most startup marketers are ignoring. Answer questions in detail, link to your canonical explainers, and participate genuinely in the communities where your buyers already are.
LinkedIn is highly cited because it is tied to real, verifiable professional identities. It signals accountability in a way that anonymous content cannot.
Medium and YouTube are both strong reinforcement channels. For YouTube specifically, structure your videos with clear titles, timestamps, and manually verified transcripts — LLMs ingest all of it.
Wikipedia is the hardest to get into, but worth pursuing if your brand has sufficient independent coverage to support a neutral entry.
The Top 10 Actions to Take in the Next 30 Days
- Run a technical audit — disable JavaScript and check what LLMs actually see
- Write your canonical brand definition using the formula above
- Create core product pages for every main feature or service
- Create “Best [category]” pages positioning your brand with data
- Create “Alternatives to [competitor]” pages for your main competitors
- Add a 150-250 word quick answer block to the top of every key page
- Add attribute-matching FAQs with schema markup to every product page
- Align your brand description across every platform — website, LinkedIn, Medium, GitHub, everywhere
- Start publishing consistently on the channels LLMs weight most: Reddit, LinkedIn, Medium, YouTube
- Write everything for LLMs — definitions, mechanics, comparisons, risks, not just narrative
Each layer builds on the one before. Technical without content is an empty site. Content without authority goes unnoticed. Authority without reinforcement fades. Build them in order and the effect compounds.
Frequently Asked Questions
What is generative engine optimisation and how is it different from SEO?
Generative engine optimisation (GEO) is the practice of making your brand visible inside AI-generated answers from tools like ChatGPT, Perplexity, and Claude. Traditional SEO focuses on ranking pages in search results for human readers to click through. GEO focuses on structured knowledge, third-party trust, and consistent brand signals so that AI systems can reliably explain, categorise, and recommend your brand without guessing.
Why can’t LLMs read most startup websites?
Most startup sites are built on JavaScript frameworks like React, Next.js, or Angular. LLMs cannot execute JavaScript, so when they attempt to read these sites they often see a blank page. The content simply does not exist for the AI. The fix is to ensure your core product content is available as plain HTML, with proper crawl directives and schema markup in place.
How long does it take to see results from GEO?
GEO results depend on which layer you are working on. Technical fixes can take effect within weeks once crawlers re-index your site. Content changes take longer — typically two to three months before LLMs begin citing new pages consistently. Authority and reinforcement are the longest-term plays, building over six to twelve months as third-party coverage and consistent brand signals accumulate across the web.
What content types does GEO require that standard content marketing does not?
GEO requires canonical explainers (“What is X?”), product pages that explicitly state what the product cannot do, comparison pages between your brand and specific alternatives, reference pages with technical parameters and limits, and negative qualification pages explaining who the product is not for. Most startup content strategies focus on announcements and thought leadership and miss these entirely.
Which channels matter most for GEO reinforcement?
Reddit, LinkedIn, Wikipedia, Medium, and YouTube are the top cited domains in LLM responses, based on Semrush data from 230,000 prompts. X (Twitter) ranks near the bottom. Startup marketing budgets that are heavily weighted towards X will see minimal return in AI search visibility terms. Redistributing effort towards Reddit, LinkedIn, and Medium produces measurably better reinforcement signals for LLMs.
Can a startup with a small marketing team realistically implement GEO?
Yes. The four-layer framework is sequential, which means you can work through it systematically rather than all at once. Start with the technical audit — it costs nothing and takes a few hours. Then write your canonical definition and apply it everywhere. Then build out the missing content types. The later layers, authority and reinforcement, benefit from automation: content repurposing pipelines, journalist outreach agents, and scheduled publishing across platforms can reduce the manual effort significantly.
Concerned about how your brand shows up inside AI search tools?
This is exactly what our LLM SEO work is designed to address. https://victoriaolsina.com/services/llm-seo-for-web3/
