SEO and GEO for startups that need signups, not just rankings.

Three things are true at once. Search Console shows impressions rising while clicks fall, because AI Overviews now answer the question on the results page. A meaningful share of buyers, especially technical ones, start their research in ChatGPT, Perplexity, or Claude, where there are no rankings, only mentions and citations. And the high-intent queries, the ones where someone is ready to buy, still run overwhelmingly through Google.

So the fundamentals of organic don't change. Some tactics do, and a few new ones matter. The mistake is treating GEO as a separate program with a separate budget. It's the same program with a wider definition of ranking.

1. 01

   Rank first; citations follow. Independent research from Seer Interactive found that page-one Google rankings correlate strongly with being mentioned by LLMs, while backlinks barely register. Most GEO is SEO done well, with a few additions.

2. 02

   Intent splits by platform and by stage. People use chat tools to understand a problem and Google to pick a vendor. Informational content has to work in the AI answer; product and solution pages have to win the click. I plan content for both, and I don't expect glossary pages to drive demos.

3. 03

   Be an entity, not a keyword. Models reason about things and their relationships. Your company needs one unambiguous description repeated everywhere: the site, the boilerplate, the directory listings, the founder's LinkedIn. "X is the Y for Z" is a sentence you should be able to find in fifty places.

4. 04

   Specific beats general. Deep pages get cited. Homepages don't. Comparisons, benchmarks, definitions, FAQs, and "how X works" pages are what models quote. Listicles and comparison pages that don't read as self-serving are the most reliable way to shape what a model says about you.

5. 05

   Recency is a retrieval signal. When a model searches before it answers, it inherits the freshness bias of the index it searches. Pages that haven't been touched in a year lose citation velocity even if they still rank. Content gets a refresh schedule, not a publish date.

6. 06

   Some of it isn't yours to control. What a model says about you is shaped partly by Reddit threads, G2 reviews, YouTube videos, and press you don't own. That's a real part of AI visibility, and it's also not what I do. I'll tell you when the gap is off-site and point you at someone who works on community or PR, rather than sell you an on-site fix for an off-site problem.

7. 07

   Measure what can be measured, and say what can't. AI referral sessions and conversions are in GA4. AI Overview wins are in the rank trackers. Share of mentions against competitors comes from prompt monitoring. None of it has search volume behind it, all of it uses synthetic prompts, and I'll tell you that on the first report instead of the last.

1. 01

   Keyword and information architecture. Before writing a word, map the site. Research the queries your buyers actually type, cluster them by theme and intent, and decide which page owns which cluster. The output is a visual map of the domain: every page assigned a target keyword profile, every gap visible, every overlap resolved. This is the step most startups skip, and it's why their blog has 300 posts and twelve pages that rank. Themes matter more than phrasing now, because models fan a single prompt out into several sub-queries and buyers arrive with seven-word questions instead of two-word keywords. How I do keyword mapping →

2. 02

   Technical. Crawl the site and fix what's broken. Confirm every page that could be cited is indexable and that AI crawlers aren't blocked at the CDN. Make sure meaningful content renders without JavaScript, because most AI retrieval doesn't wait for it. Speed. Structured data on every page that earns it: Organization, Service or Product, FAQ, Article. An llms.txt file that tells models which pages matter. A plain-text summary block high on core pages that states what you are in two sentences.

3. 03

   On-page optimization. The architecture says which pages to optimize and around what. Then the work: titles, meta descriptions, and heading structures rebuilt so each page reads as an answer to its query. FAQs added to product and service pages, including the "versus" questions buyers actually ask, with schema attached. Internal linking built deliberately across the site so authority flows to the pages that matter instead of pooling on the homepage. Images compressed, named, and given real alt text. Content structured to answer first: short sections, one idea per page, clear headings, and a reason to click through even when the answer got summarized.

4. 04

   The content engine. AI-assisted for volume, human-reviewed for quality. At Quadratic that meant scaling from about ten to 50–100 posts a month, with a person checking every piece, and every piece built around benchmark-style data, real screenshots, and an actual opinion. Older pages get refreshed on a cycle, not abandoned, because recency is a retrieval signal and a page nobody has touched in a year loses citation velocity even if it still ranks.

Use-case, integration, comparison, and template pages generated from product data can be a startup's biggest organic asset or a thin-content problem waiting to happen. At Quadratic I built a pipeline of roughly 100 steps from a SQL query to a published, schema-marked landing page, with human review gates before anything went live. The rule was simple: every page has to be the best result for its query, or it doesn't ship.

Traffic is a vanity metric with a nice chart. The number I report is organic signups, with activation and the queries that produced them. Around it, the leading indicators: impressions by theme, AI Overview presence by keyword, sessions and conversions from AI sources by landing page, and share of mentions against named competitors in the prompts your buyers actually type. If organic traffic doubles and signups don't move, we targeted the wrong queries, and I'll say so before you have to.