GEO + SEO for small shops: what actually moves the needle

Most small-shop SEO advice in 2026 is still SEO advice from 2022 with "AI" pasted into the headline. This is the actual short list of changes that move traffic for a small site, ranked by impact.

1. How do you write passages AI engines can quote?

The single highest-leverage change for both SEO and GEO is to structure your prose as a sequence of self-contained passages, each capable of standing alone as a 40 to 60-word answer to a question.

The 40-60 word range is my own rule of thumb, not a study result. The closest research is the GEO paper from Princeton, Georgia Tech, the Allen Institute for AI, and IIT Delhi (KDD 2024). The authors built GEO-bench, a benchmark of 10,000 queries, and tested rewrites like adding statistics, quotations, and cited sources. The best methods lifted visibility in generative engine responses by up to 40%, with larger gains for sources ranked lower.

What "structured" actually means in practice:

• Every major heading starts with a one-line answer: "X is...", "X refers to...", "X means..."
• The 40-60 words right after that heading are self-sufficient. A reader who lands there with zero context can understand it.
• Specific facts inside the passage: a year, a percentage, a named entity, a number.

Bad example (a real one from a startup blog I will not name):

If you've ever wondered why some websites load faster than others, the answer might surprise you. There's this amazing technology that has been around for a while now. It's changed the way we think about web performance.

Good example (same topic, restructured):

Content delivery networks (CDNs) are distributed server systems that cache and serve web content from locations geographically close to end users. A CDN reduces latency by 50-70% on average by serving assets from edge servers rather than a single origin server. The three largest CDN providers in 2025 are Cloudflare (serving approximately 20% of all websites), Amazon CloudFront, and Akamai Technologies.

The second one is 58 words. It defines the term. It has three specific facts. ChatGPT will quote that one. The first one is unquotable. That is the entire difference.

2. Which JSON-LD does each page type need?

JSON-LD is structured data embedded in a script tag that tells search and AI engines exactly what a page is: a Product, an Article, a Person, an FAQPage. Every page should carry the schema matching its type. The work is unglamorous, and it has the best cost-to-impact ratio of anything on this list.

For ToolGenX, every page type carries its own schema:

All of it is JSON-LD inside <script type="application/ld+json"> tags. Google's structured data docs recommend JSON-LD, and the AI search crawlers all parse it.

The mistake almost everyone makes is treating JSON-LD as a checkbox: "I added Article schema, done." The actual value comes from:

• Accurate dateModified and datePublished that update when content changes.
• sameAs arrays on Person and Organization linking to your real social profiles.
• @id URIs that are stable so AI engines can resolve identity across pages.
• FAQPage entries from real questions buyers ask, not invented ones.

If you are not sure how your schema looks, Google's Rich Results Test will tell you in 30 seconds. If it does not validate, AI engines will be even less forgiving.

3. Which AI crawlers should robots.txt allow?

Making AI crawlers welcome means explicitly allowing GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, Google-Extended, CCBot, and Applebot-Extended in robots.txt. A default robots.txt either allows everything, which already covers AI crawlers, or blocks some of them, which can make a shop invisible to AI search without anyone noticing. There is no neutral setting.

For a small shop that wants to be cited in AI answers, explicitly allow the major AI crawlers in robots.txt:

User-agent: GPTBot
Allow: /

User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Google-Extended
Allow: /

User-agent: CCBot
Allow: /

User-agent: Applebot-Extended
Allow: /

Two distinctions matter:

• Blocking Google-Extended stops Gemini from training on your content but, per Google's crawler docs, does not affect inclusion in Google Search (AI Overviews come from the regular Googlebot crawl).
• Blocking GPTBot stops OpenAI from training but does not remove you from ChatGPT search. OpenAI's bot docs say OAI-SearchBot surfaces sites in ChatGPT search answers, and ChatGPT-User fetches pages when a user asks. Each setting is independent.

Most small shops accidentally block themselves from one or both surfaces with a copy-pasted robots.txt. Check yours.

4. Where llms.txt fits, and where it does not

An llms.txt file at your site root gives machine readers a curated map of the site. It costs about ten minutes, it carries no downside risk, and it is the single most over-recommended item in the GEO checklist.

The short version: Google has said its Search systems do not use it, the largest public log study found most of these files are never requested, and the audience that demonstrably reads them is coding agents rather than answer engines. I ship one anyway, at agentskillpacks.com/llms.txt, because ten minutes for a maybe is a fair trade. I do not count it as a reason anyone finds us.

The full evidence, including how to check your own logs for whether yours is being read, is in does llms.txt actually work. Shipping it is fine. Shipping it instead of items 1, 2 and 3 on this list is the mistake.

5. Real author entity, real social presence

A real author entity means a named human with a verifiable public footprint: a byline, an about page, a Person schema with sameAs links. Google's quality rater guidelines added the extra E for Experience in December 2022, and raters are told to look for who created the content and why they are credible. AI search engines weigh author credibility too, because their answers carry liability.

For a small shop, this means one person, named, with a real social footprint:

• Page byline links to an /about page with full author bio.
• About page has a Person JSON-LD schema with sameAs array pointing to LinkedIn, X, YouTube, Medium, and any other real profiles.
• The same person actually posts on at least one of those platforms.
• Author photo is real, not generated.

For ToolGenX, that person is me. The Person schema on the About page lists 16 sameAs URLs. The blog byline links back. The chain is closed.

The reason this matters: when ChatGPT or Perplexity considers whether to quote a passage, the author entity is one of the trust signals it weighs. A passage by "Admin" with no further information is worth less than the same passage by a named person with a public track record.

What a small shop can safely skip

The tactics a small shop can safely skip include keyword density tuning, internal link maximization, 800-row long-tail spreadsheets, daily publishing, and manual citation chasing. Modern search is semantic and a small shop has 10 to 30 pages, so most rituals from the big-site SEO playbook cost hours and move almost nothing.

Five things people obsess about that move much less for a small shop:

1. Keyword density and TF-IDF tuning. Modern search is semantic. Write naturally about the topic and stop counting words.
2. Internal link maximization. Three to five relevant internal links per page is enough. Cramming twenty in does not help.
3. Long-tail keyword spreadsheets with 800 rows. A small shop has 10-30 pages. Pick the keywords by hand.
4. Daily content publishing. Quality + occasional updates beats volume + abandonment.
5. Manual citation chasing. Build content that gets cited because it is the best answer, then let the citations happen.

A weekend GEO sprint for a 10-page shop

A weekend GEO sprint is ten focused hours that take a 10-page shop through the full checklist. Saturday covers the page audit and rewriting the top three pages; Sunday covers llms.txt, robots.txt, the sitemap, and the author Person entity. No tooling and no subscription, just the structural fixes in priority order.

If you have a small site live today and want to apply this, here is the sequence in order:

• Saturday morning (2 hours): Audit every page with a manual checklist — does it have a self-contained 40-60 word passage at the top? Does it have correct JSON-LD? Does the byline link to a real author?
• Saturday afternoon (3 hours): Rewrite the top 3 pages to fix the structure. Add Quick Answer blocks. Update the JSON-LD.
• Sunday morning (2 hours): Write llms.txt, fix robots.txt, submit updated sitemap.xml to Google Search Console.
• Sunday afternoon (3 hours): Build out the author Person entity on About, link sameAs, fill in real social URLs.

That is one weekend. By the following week, AI engines will have re-crawled. By month two, you will have data on whether citation share is moving.

I have run this sprint for myself and for two friends. It works.

Tools that automate parts of this exist. AI Search Visibility Toolkit is the GEO audit + scoring system I use myself; Structured Data Pro Pack is the JSON-LD library that powers ToolGenX's own schema. Both are products I built to solve my own problems first.