SEO for AI (GEO/AIO): 4-step process to find, optimize and scale AI-cited content

This article is not for new websites, but for sites that have been doing SEO for a while and start to get traffic from AI (ChatGPT, Gemini, Perplexity, Copilot, AI Or…). The goal is: discover which URLs are being cited by AI → dissect the reasons → optimize systematically → scale across the site.

Background thinking: How is “AIO/GEO” different from traditional SEO?

• Traditional SEO: optimized for wins on web SERP (10 blue links, PAA, FS…).

• AIO/GEO (Answer/Generative Engine Optimization): optimal so that AI chooses you as a citation source in the summary answer.

• Intersection point: Content has good E-E-A-T + clear structure + answers the right intent and helps “Top 0/Featured Snippets”, has just increased the probability of being cited by AI.

Process overview (4 steps)

1. Export data URLs with traffic from AI (GA4, Looker Studio)

2. Filter & prioritize: URLs cited by many AIs / URLs with large traffic from one AI

3. Ask/search on LLMs: see how AI displays & quotes you like and why

4. Audit SEO & content optimization → standardize into a checklist → scale goods series

Step 1 — Export all URLs with traffic from AI

1.1 Quick filtering in GA4 (fast, relatively new data)

• Navigation: Reports → Life cycle → Engagement → Landing page

• Add filter: Session manual source / Session source

Many platforms/LLM attach UTM source when users click on the original link. You can expand the search using the “contains” filter for upper/lower case variations.

1.2 Build a Dashboard in Looker Studio (bulk export)

(The correct name is Looker Studio, easily mistakenly written as “Locker Studio”)

• Connect to GA4 → create 1 report page new → Table

• Dimension: Landing page (or Page path + query string if full URL is needed)

• Metrics: Views, Users, Sessions…

• Filter: Session source → contains → chatgpt (create multiple filters for each AI)

• Create multiple tables (one table for each AI) or a common table with subdimension is Session source.

• Export CSV after selecting last 30 days, back 2 days compared to the current one to reduce data latency.

Suggested list of common “source”/“medium”

chatgpt, gpt, openai, gemini, perplexity, copilot, bing-ai, ai-hay, ai_hay

Actual UTM may vary by user/application. Open “Source/Medium” unfiltered first to check for variations.

Step 2 — Filter data & choose optimal priority

Export data to Excel/Google Sheets and create criteria:

Suggested set of criteria

1. “Multi-AI”: URL appears at ≥ 2 AI sources (for example: both ChatGPT and Perplexity). → This is a “good pattern”, worth dissecting and scaling.

2. “Single AI dominates”: URL with high traffic from one AI (e.g. only from ChatGPT). → Optimize in depth for that AI.

3. New growth: URL newly started getting traffic from AI in the last 30 days. → Prioritize “nurture”.

Suggested column table in Sheets

URL | Sessions (ChatGPT) | Sessions (Gemini) | Sessions (Perplexity) |
Sessions (Copilot) | Sessions (AI Good) | Total Sessions AI | #AI sources | Priority Tier

Sample formula (Google Sheets)

• #AI sources: count columns sessions > 0

  =COUNTIF(B2:F2, ">0")
  

• Priority Tier (logical example):

  =IF(H2>=3,"Tier 1",
     IF(AND(H2=2, G2>=100),"Tier 1",
        IF(G2>=50,"Tier 2","Tier 3")))
  

Step 3 — Ask / search on LLMs to see how AI shows & quotes you do

3.1 Get “seed keywords” to interact with AI

• From Ahrefs/Semrush/GSC, get:

  • Keywords for which the URL is ranking 1–10

  • Queries with Featured Snippets

  • Long-tail “what/why/how/who/where/when” (tends to trigger AIO)

• Choose 5–10 queries that represent each URL.

3.2 Prompt template (used when asking LLM)

Goal: recreate real user context + force AI to yes/no quote.

Template 1 — Simulate real query

You as a user need short, precise, quoted answers source. 
Question: [Your “what/why/how” query]. 
Please answer in the following format:
1) Conclusion in 2–3 sentences
2) Bulleted list of key points (3–7 bullets)
3) Reference source with URL

Template 2 — Check “source”

With query: “[keyword]”,
- List 5–10 sources you most often cite in your answer.
- For each source, state the reason you chose it (reputation, completeness, clear structure, original data...).

Template 3 — See how AI “picks up” content

I provide the URL: [Your URL].
- Please extract the pieces of content in this URL that you consider most useful for the question “[keyword]”.
- Suggest which structures I should edit/add to increase my chances of being quoted by you.

Experience: Featured Snippets and SERP wins over “summary/table/list” are highly correlated with AIO (independent studies have documented significant similarity between traditional top results and sources appearing in AI responses). Take advantage of this to choose the right presentation format.

3.3 What to write down

• WHO has/doesn't cite? Quote Which URL (you/competitor)?

• Presentation format used: table, timeline, table of contents, bullet, step-by-step…

• What piece of the puzzle is missing compared to your competitors? (definition, number, example, disclaimer, schema...)

• Which “conversation language” is triggering AI to display (sub-questions, Q&A in the article)?

Step 4 — Audit SEO, content optimization & mass scale4.1 “AIO-first” audit framework

A. Structure & presentation (so AI can easily quote)

• TL;DR / Executive summary 3–5 sentences at the beginning.

• Table of contents is clear, H2/H3 is consistent; each H2 solves 1 concept.

• Table/Checklist where appropriate (AI loves to carry tables).

• Q&A block for conversational queries (what/why/how/who…).

• Short illustrative examples (cases, numbers, templates).

B. Data & evidence

• Insert standard definitions, numbers, reputable sources (selective out links).

• Schema appropriate: Article, FAQ, HowTo, Product/Service, Organization, Person (if there is an author).

• Update modified date (Last updated) and reason for update (short changelog).

C. E-E-A-T & trust

D. Internal link by cluster

• From satellite article → pillar article and vice versa.

• Anchor natural, but semantic match.

Refer to the foundational playbook on topic clusters:
https://tanphatdigital.com/vi/resources/seo-guide

4.2 Optimal demo (from the case “celebrity biography”)

• Biography table: real name, year of birth, occupation, “famous star” column split bullet details (for example: “transition effects”, “catchy music”, “style…”) → AI is easy to pick up.

• Career Timeline: milestones, products/awards → table format/timeline.

• Q&A: “Who is A?”, “Famous for what?”, “What drama?”, “Achievements main?”.

• Schema: Person + FAQ.

• Outbound link: official source (press/awards) → increase credibility.

4.3 Standardize checklist & workflow to scale

Create checklist AIO-first for the editing team:

Writing/editing checklist (shortened)

• TL;DR 3–5 sentences

• Automatic table of contents

• 1–2 useful tables/templates

• 3–7 bullet key takeaways

• Q&A 4–6 what/why/how/who questions

• Appropriate schema + updates dateModified

• Internal 2-way link in cluster

• 2–3 trusted external sources

Workflow 90–120 minutes/lesson (suggested) ideas)

1. 10’ SERP + AIO review (template asks LLM)

2. 20’ final outline + table/Q&A required

3. 40–60’ content writing/editing

4. 10’ schema + internal links

5. 10’ QA (E-E-A-T, spelling errors, alt text)

4.4 Measurement – KPI “AIO/GEO”

• AIO Impressions (estimated): number of times you appear in AI answers (difficult to measure directly → use proxy as sessions from AI source).

• Sessions from each AI by URL/topic cluster (GA4/Looker).

• #AI sources/URL (multi-source → expansion priority).

• FS coverage: % of queries with Featured Snippets that you own.

• SERP CTR: improve title/meta relative to exported AIO existing.

• Time-to-update: SLA from discovery → optimization → re-crawl.

Some "bloody" notes

• Don't over-audit: A/B test in 1 small cluster before scaling.

• Maintain E-E-A-T: AI "trusts" sites with real credibility.

• Don't spam Q&A: keep it natural, serve readers first.

• Focus on "winning patterns": The URL has been cited by many AIs as a guideline for the next publishing template according to.

• Featured Snippets ≈ AIO: optimization for FS (short definitions, steps, tables…) often helps increase the possibility of being “chosen by AI”.

AIO/GEO does not replace traditional SEO – it inherits and adds a layer of optimization to the answer AI. When you know which URLs are being cited by the AI, understand why the AI chooses (structure, tables, Q&A, E-E-A-T) and turn that into a mass publishing checklist, you will unlock new growth levers that many competitors have not yet caught up with.
The core remains: useful content, well-structured, trustworthy – easy to cite guide. The rest is measurement discipline and iteration rate. With that mindset, the team can confidently scale GEO sustainably.