Field guide · September 2026
AEO & GEO: get found, cited, and recommended by AI.
A hands-on system for improving how your company appears in ChatGPT, Google AI experiences, Perplexity, Gemini, Copilot, and the search systems underneath them.
The premise
“How do I rank in ChatGPT?” is useful shorthand, but the wrong mental model.
There is no single permanent LLM position to win. Answers can change by query, engine, retrieval results, location, account context, and time. The practical goal is to increase the probability that systems retrieve you, understand you correctly, cite useful material from you, and recommend you for the questions that matter.
SEO
Get discovered
Crawlability, indexing, relevance, internal links, authority, useful pages.
AEO
Get extracted
Direct answers, definitions, evidence, structure, useful machine-readable context.
GEO
Get cited or recommended
Entity clarity, source-worthiness, corroboration, retrieval coverage, accurate portrayal.
Do not abandon SEO. Search visibility, answer extraction, and generated recommendations are connected systems. Measure each layer separately so you know what actually changed.
Workflow 01 · Query universe
Decide what you want to be visible for.
Start with customer questions, not a GEO tool. Include problems, definitions, categories, workflows, comparisons, alternatives, recommendations, and branded questions. One conversational question can expand into several related searches and sources.
Starter query set
- 01What is the best design context for AI coding agents?
- 02How do developers building interfaces with AI coding tools keep AI-generated UI visually consistent?
- 03What should I give an AI coding agent so it follows my design system?
- 04What is DESIGN.md and how does it work?
- 05Best tools for design context for AI coding agents
- 06DesignMD.co alternatives
- 07Is DesignMD.co useful for Claude Code or Cursor?
- 08DesignMD.co vs writing a design system prompt manually
- 09How do I turn an existing website into design context for an AI agent?
- 10What tools help AI coding agents understand typography, spacing, color, and components?
Workflow 02 · Baseline
Measure before you optimize.
Run the same important questions across the answer surfaces your audience uses. Track presence, prominence, portrayal, citations, competitors, and factual errors. The point is a repeatable baseline, not a vanity score.
Baseline protocol
Repeat the protocol later. Consistency matters more than one vendor score.
0/7 checked
Benchmark spreadsheet columns
Paste this tab-separated header into Sheets, Excel, Airtable, or Notion.
Live case study · DesignMD.co
Turn search data into an intelligence layer, not a screenshot of metrics.
DesignMD.co gives us a real starting point. The goal is not to imply that Google performance equals GEO performance. It is to use existing search behavior to identify entities, demand, opportunity pages, audience patterns, and anomalies before building the separate AI visibility benchmark.
60.3K
Google clicks
529K
Google impressions
11.4%
blended CTR
Last 3 months
export window
Signal 01 · Entity demand
People already search for the entity by name.
The three leading branded variants, “design md”, “design.md”, and “designmd”, rank around positions 2.6 to 3.0 in this export, with CTRs of roughly 23% to 32%. That is evidence of branded search demand and a relatively clear search entity. It is not evidence, by itself, of LLM visibility.
Signal 02 · Page portfolio
Entity pages create measurable discovery surfaces.
DesignMD.co has indexed pages for recognizable design systems and brands. Instead of saying “programmatic pages work,” use Search Console to see which entities have demand, which pages already earn visibility, and where improvement work is most justified.
Signal 03 · Device mix
Search behavior is overwhelmingly desktop-led.
Desktop contributes about 89% of clicks in the export. That should influence QA and conversion analysis, but it does not mean mobile should be ignored.
Analyst note
Do not turn every large number into a success story.
One DesignMD.co URL in the export has about 122K impressions but only 0.15% CTR. That is an anomaly to investigate, not a metric to celebrate or “optimize” blindly. First inspect the query mix, intent mismatch, indexing, and whether the page is appearing for an unrelated term.
The BI lesson
A dashboard should move from metric → signal → diagnosis → action → retest. GEO measurement should work the same way.
Generative AI layer
Read generative search as a separate intelligence layer.
Google Search Console now exposes a generative AI view. For DesignMD.co, the useful question is not whether the raw volume is “large.” It is what Google’s generative surfaces are selecting, how broadly that retrieval is distributed, and what to improve next.
20,476
generative AI impressions
Export window: May 18 through September 5, 2026.
What Google is selecting
The footprint extends beyond the homepage.
The homepage contributes about 43% of generative impressions. The catalog’s individual /d/ pages collectively contribute roughly 10.4K, with retrieval spanning entities such as Apple, Anthropic, Ramp, Linear, Duolingo, Revolut, Lovable, Cloudflare, Cursor, Microsoft, Airbnb, Claude, and Perplexity. The practical signal is distributed entity retrieval, not just branded homepage exposure.
Trajectory
Watch persistence and breadth.
~42
first 28 days / day
~378
recent 28 days / day
Use this as a trajectory signal, not proof that a specific optimization caused the change. The recent 14-day period was also about 11% above the prior 14 days.
Device pattern
Generative visibility is desktop-led too.
Geographic breadth
The retrieval footprint is international.
The United States leads the export, while Indonesia, India, South Korea, Brazil, Germany, Thailand, the United Kingdom, Japan, and Vietnam also contribute meaningful generative visibility.
For a globally relevant software resource, country distribution can reveal where the entity model and content travel without assuming every market has the same intent.
How to spin the data responsibly
Observe
Which entities, pages, countries, and content types are already being selected?
Diagnose
Why might those surfaces be useful to a generative result? What is missing from adjacent pages?
Extend
Improve entity clarity, answer coverage, evidence, internal links, and source-worthy editorial content, then retest.
Three measurement layers
Traditional Search: clicks, impressions, CTR, position.
Google Generative Search: generative impressions, pages, geography, devices, trajectory.
LLM benchmark: mentions, recommendations, citations, portrayal across AI answer engines.
Do not collapse these into one opaque “AI visibility score.” Each layer answers a different business question.
01
Observe
Which entities and questions already create search demand?
02
Prioritize
Which pages have enough visibility to justify better answers, evidence, and intent alignment?
03
Extend
Use those insights to build the LLM query set, then measure AI citations and recommendations separately.
Workflow 03 · Retrieval
Make sure the systems can reach the useful page.
Audit robots.txt, noindex, canonicals, authentication, CDN or WAF rules, internal links, sitemaps, and whether important information exists in accessible text.
Googlebot
AI Overviews and AI Mode use Google's Search systems. There is no separate GEO crawler or special AI schema requirement.
ChatGPT Search
OAI-SearchBot
Allow OAI-SearchBot if you want public pages available for ChatGPT Search summaries, snippets, citations, and links.
Bing / Copilot
Bingbot + IndexNow
Verify Bing Webmaster Tools and use IndexNow when important URLs are added, updated, or removed.
Perplexity
PerplexityBot
Check robots.txt and your CDN or firewall if you want Perplexity to retrieve public pages.
llms.txt: treat it as an experimental complement, not a ranking switch.
Workflow 04 · AEO
Make the answer easy to extract without writing for robots.
A strong answer page gets to the point, supports claims, explains nuance, and anticipates the next question. This is useful for people first, which is why it is also useful to retrieval systems.
The Answer Block
Use this as a page-editing checklist, not a rigid SEO template. The goal is to make the useful answer easy for a person and a retrieval system to find.
QUESTION What is [topic]? DIRECT ANSWER Answer in 2-4 sentences. Define the thing before selling it. EVIDENCE Add primary sources, original data, examples, dates, or methodology near the claim they support. EXPLANATION Explain how it works and when it is useful. LIMITS State important exceptions, tradeoffs, or uncertainty. NEXT QUESTIONS Answer the natural follow-up questions on the same page. SOURCES Link to the strongest primary evidence.
Workflow 05 · Citation engineering
Do not publish another commodity blog post. Create a source.
Original datasets, benchmarks, methods, definitions, calculators, public tools, comparison research, and useful primary documentation give other people and systems a reason to cite you.
Run your important queries
Save every domain and page the answer engines cite, not just the brands they mention.
Group the sources
Separate publishers, forums, directories, research, documentation, reviews, videos, and first-party sites.
Ask why each source wins
Look for original data, clear definitions, comparisons, reputation, freshness, firsthand experience, or unique utility.
Find the gap
Decide whether you need a better page, a better source asset, a legitimate third-party mention, or clearer entity information.
Create something worth citing
Publish a benchmark, dataset, methodology, calculator, definition, comparison, public tool, research note, or genuinely useful guide.
Tool stack
Use a tool for a job, not because it has a GEO label.
First-party data tells you what the platform can actually report. Manual benchmarks show the answer itself. Third-party tools help you scale monitoring. None of them replaces diagnosis.
First-party
Google Search Console
Use conventional Search data plus Google's dedicated generative AI performance reporting. Start here before buying another dashboard.
First-party
Bing Webmaster Tools
Search diagnostics and Microsoft's AI visibility reporting where available.
Manual benchmark
ChatGPT, Google AI, Perplexity, Gemini, Copilot
Run a stable query set and record mentions, citations, competitors, portrayal, and errors.
Technical
Rich Results Test + Schema.org
Validate structured data that accurately describes visible page content. Do not invent AI-only markup.
AI visibility
Profound / Scrunch
Higher-end monitoring and analysis for teams that need broader prompt and citation coverage.
AI visibility
Peec AI / Otterly / SE Ranking / Semrush
Ongoing prompt, brand, and citation monitoring. Compare methodology and coverage, not headline scores.
30-day operating system
Measure → retrieve → answer → earn authority → retest.
Week 1
Measure
Define 30 to 50 questions. Benchmark the engines that matter. Record citations, competitors, portrayal, and errors.
Week 2
Fix retrieval
Resolve crawl and index issues, improve internal links, align entity facts, validate structured data, and expose important information as text.
Week 3
Improve answers
Upgrade five high-value pages with direct answers, evidence, useful comparisons, limitations, and related questions.
Week 4
Build authority
Study citation sources, publish one source-worthy asset, earn legitimate distribution, then rerun the same benchmark.
Keep this distinction
Search Console is evidence about search. Your AI benchmark is evidence about AI answers.
Use them together, but do not collapse them into one story. The useful question is whether improvements to retrieval, entity clarity, answer quality, and authority change the visibility outcomes you measure over time.
Build alongside
other marketers.
Ask questions, share what you're building, and get unstuck faster. The community moves at the pace of people who are actually shipping.
Ask anything
No question too basic
Share your builds
Real feedback from peers
Early access
New guides first
Free to join · No credit card · Unsubscribe any time