How to Build a GEO Strategy Without an Agency or Tools

Can You Actually Do GEO Without Tools or an Agency?

Yes — the fundamentals of Generative Engine Optimization are documented, the primary research is public, and the measurement can be done with a spreadsheet and an hour a week. What you can't do without tools or help is scale it — and knowing where that wall is matters more than pretending it doesn't exist.

This is the honest playbook: what to do yourself, in what order, and the exact point where DIY stops being a strategy and becomes a hobby.

TL;DR: Six steps, all free — baseline your AI presence (1 hour), fix crawler/indexing basics, restructure top pages answer-first, deploy schema + llms.txt + Wikidata, build third-party presence on Reddit/review platforms, and track 20 queries weekly in a spreadsheet. Cost: 4–6 hours/week. It breaks at ~10 pages or multiple clients.

Step 1: Baseline Where You Stand (Free, 1 Hour)

Before optimizing anything, measure what the engines already say about you.

• Query the engines directly. Write down 20 questions your buyers actually ask — "best [category] for [use case]", "is [your brand] good", "[competitor] vs [you]". Run each in ChatGPT, Perplexity, Claude, and Google. Document: mentioned? cited? accurate?
• Check accuracy, not just presence. 78% of brands have factual errors in how AI engines describe them — being mentioned wrongly is worse than absence.
• Log it in a spreadsheet. Query, engine, date, mentioned Y/N, cited Y/N, who got cited instead. This becomes your baseline every future measurement compares against.

That afternoon tells you which of the two problems you have: invisible (engines don't know you — entity problem) or uncited (they know you but can't extract you — structure problem). The fixes are different. Full framework: AI visibility tracking.

Step 2: Fix the Retrieval Basics (Free, Binary)

The "car won't start" tier — nothing below matters until these pass:

• robots.txt allows AI crawlers: GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended
• Indexed in Bing — ChatGPT and Perplexity both lean on it. Submit your sitemap to Bing Webmaster Tools
• Content renders server-side — view-source your key pages; if the text isn't there, neither are you
• Canonicals and sitemap submitted to Google Search Console

Step 3: Restructure Your Top Pages (Free, Highest Leverage)

The Princeton GEO study — the founding research — measured it directly: statistics lifted visibility +41%, source citations +28%, keyword stuffing below baseline. Structure is the mechanic:

• Answer-first lead. Every priority page opens with a 40–60 word direct answer — liftable verbatim
• Question-based headings. The buyer's actual question as the H2/H3; the answer directly under it (2.8x citation-rate correlation)
• FAQ block on every key page — it maps to how engines extract
• Self-contained chunks. Each section readable alone, because that's how it gets quoted
• Real citations. Link every claim to a primary source — this is the +28% lever, and it costs nothing

Step 4: Machine-Readable Identity (Free, One Afternoon)

• Organization schema — JSON-LD with name, url, logo, sameAs, founder. Validate in Rich Results Test
• Article/FAQPage schema on content — table stakes (setup guide)
• llms.txt at your root — ~32% coverage lift in 14 days in Erlin's data. Thirty minutes, our build guide
• Wikidata entry — the highest-ROI entity build that exists (entity SEO)

Step 5: Third-Party Presence (Free-ish, Slowest)

The earned-media bias is real — engines weight what others say about you over what you say (Ahrefs' 75K-brand analysis: mentions 0.664 vs backlinks 0.218). Free channels that move it:

• Reddit — genuine helpful presence in your niche's subreddits (3.4x citation lift); astroturfing gets detected, so actually help
• Review platforms — G2, Capterra, relevant directories (2.6x lift)
• Guest expertise — quotes in others' content, podcast appearances, community answers — every one is a grounding signal

Step 6: Measure Weekly (Free, The Discipline)

Twenty queries, four engines, every week, logged. That's the whole measurement stack. Coverage, mention rate, citation rate, share of voice — the tracking guide has the spreadsheet format. The trend over 90 days is the signal; weekly snapshots are noise.

Where DIY Hits the Wall

Honest math: this playbook runs about 4–6 hours a week done properly — baseline queries, weekly tracking, quarterly refreshes, ongoing entity work. That's sustainable for one site. It breaks when:

• You need it across 10+ pages × 4 engines × multiple clients/brands — the tracking alone becomes a job
• Production volume is the bottleneck — you know what to write and can't produce it weekly
• Freshness decay outpaces you — content under 3 months old is up to 3x more likely to be cited, so stopping is regressing

That's the exact point where the three paid paths open — tools, hire, agency, or owned infrastructure. The honest comparison: agency vs. tools vs. in-house.

Frequently Asked Questions

Can I do GEO completely free?

Yes — every step above is free. The cost is 4–6 hours/week of disciplined execution. Free strategy, paid-in-time execution.

How long until DIY GEO shows results?

Structural fixes (answer-first, schema, llms.txt) show movement in 2–4 weeks; machine-readable formats measured lift in 14–21 days. Entity authority takes quarters. Anyone promising faster is selling the shortcut, not the work.

What's the single highest-ROI free action?

Rewriting your top page to lead with a 40–60 word direct answer under a question-shaped heading — combined with an llms.txt file. Combined cost: one afternoon.

When should I stop DIY and pay for help?

When the weekly tracking cadence slips twice, when production can't keep pace with the refresh schedule, or when you're managing more than ~10 priority pages. At that point the constraint is capacity, not knowledge.

DIY'd the strategy and hit the capacity wall? The Altyzo platform automates the execution for agencies — $30K license. For brands: Altyzo Managed runs it for you.