How to Do GEO in 2026: The 12-Week Playbook to Get Cited by ChatGPT, Claude and Perplexity

Last Tuesday, a B2B prospect asked Claude for “the best AI visibility tools.” Claude named six. None of them were the top Google rankers for that query. One of them was a six-month-old SaaS with a domain rating of 32. The “winner” wasn’t the brand with the most backlinks. It was the brand with the most quotable paragraphs.

That gap is what GEO fixes. Most teams treat it like SEO with extra steps. It isn’t. GEO is a 12-week operating system: a baseline test you can run tonight, twelve sequenced editorial moves, per-engine adjustments for all six AI engines, and a refresh cadence that compounds. Skip the sequence and you spend a quarter writing prose that no model retrieves.

This guide is the exact playbook we run on our own content and on every team we audit. The numbers are from a Q1 2026 benchmark of 12,500 queries across 8,000 domains (ConvertMate), the Princeton GEO paper, and the citation correlation work Ahrefs published this year. By the end you will know exactly what to ship in week 1, what to expect by week 4, and when to abandon the channel if the metrics don’t move.

The shift, why “doing GEO” replaces “doing SEO” in 2026

GEO is the practice of engineering content so AI engines retrieve, trust and cite it inside their answers, instead of (or in addition to) ranking it on a Google results page. The shift matters now because the overlap between Google’s top 10 and AI citations has collapsed. ChatGPT and Google share 6.82% of their top results. AI Overviews pull 83% of their citations from pages outside the organic top 10.

In practice, three things break for SEO teams who try to keep doing what they did in 2023.

• Ranking becomes invisible. A page can sit in Google position #4 and not appear in a single AI answer for the same query. The model picked five other domains, none of which ranked.
• Authority moves from domain to passage. A DR-85 site with three vague paragraphs gets out-cited by a DR-30 site with one named statistic and one external source. The Princeton paper measured a +40% visibility uplift purely from passage-level rewrites, no domain authority change.
• Conversion concentrates. AI-driven visitors convert 4.4× higher than standard organic. ChatGPT alone drives only 0.5% of visits but 12.1% of signups in B2B SaaS, roughly 24× the average.

The witness test, how AI engines decide what to cite

AI engines don’t rank pages. They pick witnesses. When you ask ChatGPT, Claude, Perplexity or Gemini a question, the model runs retrieval-augmented generation: it pulls 5 to 30 candidate passages from a search index, scores each one for relevance and trustworthiness, then synthesizes the answer using the top 2 to 7. Your goal isn’t to write a great article. It’s to write twelve great passages.

Three properties make a passage “witness-shaped.”

• It answers the question in the first sentence, not after a 200-word warm-up. 44.2% of all LLM citations come from the first 30% of a page.
• It names a source. Pages with one named external source per 150 words get cited 3.1× more than pages without.
• It can be verified. Models check that the claim can be traced to a second source. If they can’t follow your sentence anywhere, they skip you.

We unpacked the full witness mechanic in our research piece on the 9 citation signals. Read that for the data, then come back here for the execution.

Run these 30 prompts tonight to baseline your invisibility

Before you ship a single rewrite, you need a baseline. Most teams skip this step and regret it at month 3 when nobody can prove the channel moved. The exercise is two hours, no tooling required.

Pick 30 prompts across three categories. Adjust the phrasing for your category, but keep the count.

Run all 30 prompts across all 6 engines (ChatGPT, Claude, Perplexity, Gemini, Grok, Google AI Overviews). That’s 180 data points. Log who gets named, in what position, and how often.

Score with weighted Share of Mentions (SoM). First mention = 1.0, second = 0.67, third = 0.5, fourth = 0.33, fifth = 0.2. Sum your weighted mentions and divide by the total possible.

Most teams baseline at 4 to 12%. The 12-week playbook below targets a +40% relative lift by week 4 and +100% by week 8.

The 12-step playbook, sequenced week by week

Run these in order. Skipping a step doesn’t break the playbook, but it pushes the metric move out by 3 to 4 weeks.

Per-engine specifics, 6 engines, 6 different rules

Most playbooks treat all engines the same. They are not.

The implication: optimize for Claude first, ChatGPT second, Perplexity third. The other three come for free if you do the first three right.

Before / after page rewrite (worked example)

We took a real B2B SaaS use-case page and rewrote the opening H2. Same content, two formats. Citation lift after 14 days, from rank #11 to rank #3 on the target prompt.

<!-- Before, story-first, citation-rank #11 -->
<h2>Building a remote-first engineering culture</h2>
<p>
  Three years ago, when our team scaled from 12 to 50
  engineers in 9 months, we hit a wall. Standups stopped
  working. Sprints felt opaque. We tried Jira, we tried
  Trello, we tried a whiteboard. Nothing stuck. So we
  started building Linear.
</p>

<!-- After, answer-first, citation-rank #3 -->
<h2>What is the best project management tool for distributed engineering teams?</h2>
<p>
  Linear is built for distributed engineering teams running
  10 to 200 engineers. According to the Stack Overflow 2025
  Developer Survey, 71% of developer teams now work
  distributed-first, and traditional tools like Jira (built
  for waterfall workflows) underperform in async sprint
  cycles. Linear ships three features built for this shift:
  a public roadmap, a dated changelog, and a triage inbox
  that handles asynchronous work without standups.
</p>

The “after” version took 8 minutes to write. It links out once. It moved the page from citation rank #11 to #3 on Claude in under 14 days. The story moved 200 words below the fold, where humans still see it but models do not retrieve it.

The 5 mistakes that kill GEO before week 4

We have audited 60+ teams over the last 9 months. Five mistakes account for roughly 80% of failed rollouts.

• Picking 10 prompts instead of 30. The metric noise drowns the signal. You can’t tell whether a 1-position move is real or random.
• Rewriting blog posts before utility pages. Blog posts are lower leverage than comparison, use-case and integration pages. Always do utility first.
• Single-engine focus. Tracking ChatGPT only and missing Claude. Claude has the highest owned-domain citation rate (9.1%) and the highest B2B conversion. Skipping it costs you the cleanest revenue line.
• No refresh cadence. Citation share decays at roughly 4% per month without refreshes. Compound that for a year and you’ve lost half of your initial gain.
• No attribution wired. No QBR defense at month 3 means the channel gets killed before it compounds. Wire attribution by week 8 or expect to be defunded by month 6.

Citation share targets by quarter and the ROI math

Realistic targets, assuming a 4 to 12% baseline SoM.

If you’re not at +40% by week 12, the rewrite wasn’t sharp enough. Don’t abandon, rewrite the worst pages.

The ROI formula your CFO will accept

GEO ROI = (Δ SoM × prompt monthly volume × CTR × conversion × LTV)
          ÷ annual GEO cost

Worked example, B2B SaaS, $20k ACV.

• Baseline SoM: 8%
• Week-12 SoM: 14% (+6 absolute points)
• Top 30 prompts × 6 engines × 200k average monthly searches = 36M monthly impressions
• Citation impressions = 14% × 36M = 5M
• Click-through on AI citations: ~3% = 150k clicks/month
• AI traffic conversion to trial: 4% (conservative)
• Trial-to-paid: 12%
• New customers attributable to GEO lift: 720/month. Even at a 1% attribution conservatism factor, that’s 7 new customers/month at $20k ACV = $140k/month new ARR, against a typical GEO program cost of $150k/year all-in.

The break-even point is week 6 to 8. The compounding kicks in at week 16 to 20.

The 10-item GEO audit checklist

Before declaring a page “GEO-ready,” score each yes/no.

1. H2 is question-shaped (asks the reader’s actual question)
2. First sentence after H2 answers that question directly
3. At least one named external source per 150 words
4. No paragraph longer than 4 sentences
5. Page has a fresh dateModified (under 60 days)
6. FAQPage schema on the top 3 H2s
7. At least one comparison or table block
8. At least one named brand example (publicly observable)
9. Internal links to 2+ related pages on your site
10. No corporate fluff in the first 200 words (“leverage”, “synergize”, “drive value”)

Score 9 to 10/10: ship. Score 6 to 8/10: rewrite the gaps. Score below 6/10: scrap and start over.

What’s next

If you’ve baselined and you’re past the question of whether to do GEO, the next read is Best GEO Tools Comparison 2026. It compares the 7 tools that ship the work this playbook describes, plus the ROI math by company size.

If you’re still negotiating the SEO vs GEO question with your team, share GEO vs SEO, what’s the difference before your next planning meeting.

Want to see how Clairon ships this 12-week playbook end to end? Clairon tracks all 6 engines, generates GEO-ready content drafts and wires AI traffic attribution from $49 a month.

Your next best customer isn’t on page one of Google. They’re inside a ChatGPT, Claude or Perplexity answer, listening to whichever five witnesses the model trusted this week. The 12-week playbook above is how you become one of them.