Classroom Session · 90 min

Search Optimisation
for AI Chatbots

How to become the source AI systems quote, summarise, and trust.

Rudranil Chakrabortty

Instructor, Funnelysis

Warm-up

Lets Analyze this Website

Look at it like a search engine would. What's on the page? Who wrote it? What questions could it answer?

krantivo.serverside.in

Open in your browser →

SEO

SEO still matters.

Two reasons traditional search optimisation is the foundation — not the obsolete predecessor — of AI search.

Reason 01

Keywords didn't die — they got more context. Users moved from typing keywords to asking questions. But every question is still built around the same core terms.

Before: landing page builder pricing Now: "What's the best landing page builder for a small team with a limited budget?"

The intent is the same. The context around it is richer.

Reason 02

Live AI still runs through search. For real-time queries — prices, news, fresh comparisons — chatbots don't answer from memory. They query a search engine, retrieve pages, and synthesise.

If you're not indexed and rankable, you're not retrievable.

SEO is the entry ticket. Without it, you're not even in the room.

Mental Model

Two different optimisation games.

Traditional SEO	AI Search Optimisation
Rank pages	Get cited, mentioned, or summarised
Keyword → page	Question → answer fragment
CTR-focused	Visibility + trust + citation-focused
Page-level optimisation	Passage, entity, and source-level optimisation

In SEO, you optimise for the crawler and the ranker. In AI search, you optimise for the retriever, the ranker, the summariser, and the citation layer.

AI Mechanics

How AI chatbots actually answer.

Two fundamental techniques — and the technique determines your optimisation strategy.

Mode 01 · Training Data

The model learned from a large snapshot of the web before launch. When you ask it something, it draws on that baked-in knowledge.

No live search. No retrieval. Just what it already knows.

What was published, cited, and discussed at scale before the cutoff is what gets embedded.

For your brand: if you barely existed on the public web before the cutoff, the model doesn't know you — or gets you wrong. You can't fix this retroactively. You can build for the next training cycle.

Mode 02 · RAG

Retrieval-Augmented Generation. The model fetches content from the web at the moment of your query, then constructs an answer using what it just found.

Live. Fresh. Source-dependent.

Most chatbots use a combination of both — which one drives a given answer changes your optimisation strategy entirely.

For your brand: RAG is where you have direct leverage. The retrieval, ranking, and citation steps are all influenced by what's on your page.

RAG

The retrieval flow — six steps.

User asks

A natural-language question.

Interpret intent

Break into retrievable concepts.

Retrieve

Via Bing, CommonCrawl, or a proprietary crawler.

Select passages

Score by relevance & credibility.

LLM synthesises

Writes an answer from the passages.

May cite

Perplexity always. ChatGPT & Gemini sometimes.

Best teaching example

Citations are visible in the UI on Perplexity — you can watch exactly which sources got chosen, in what order, and infer why.

Spend 20 min/week running category queries there. You'll learn more about how AI ranks sources than any blog post can teach.

Systems using RAG today

ChatGPT with browsing
Perplexity by default
Microsoft Copilot
Gemini with Search
Google AI Overviews

Crawlers

Who's visiting your site.

Crawler	Platform	What it feeds
GPTBot	OpenAI / ChatGPT	Training data + live retrieval
ClaudeBot	Anthropic / Claude	Training data
Claude-User	Anthropic / Claude	Live URL fetch on user request
Claude-SearchBot	Anthropic / Claude	Search indexing
PerplexityBot	Perplexity	Live retrieval
Bingbot	Microsoft / Copilot	Index powering multiple AI tools
CommonCrawl	Open dataset	Training data for many LLMs

Each can be allowed or blocked independently in your robots.txt. Default recommendation: let them all in. If you're blocking AI crawlers without a specific reason, you're opting out of AI search visibility by accident.

Round 2

Let's analyse another website.

Now with the right vocabulary. Is it crawlable? Does it answer questions? Does it have schema? Named author?

flowmira.serverside.in

Open in your browser →

AEO + GEO

Two disciplines, one goal.

AEO · since 2017–18

Answer Engine Optimisation. Emerged with voice search and featured snippets.

Focus: structure content so it can be extracted and delivered as a direct answer — by Google's answer boxes, Alexa, Siri, or any system looking for a definitive response.

GEO · since 2023

Generative Engine Optimisation. Newer term, gaining traction with mainstream adoption of ChatGPT.

Focus: optimise for systems that don't just extract an answer but synthesise one — pulling from multiple sources, paraphrasing, choosing which to attribute.

AEO is about being the answer. GEO is about being a trusted source in a synthesised answer.

AEO

Answer Engine Optimisation — how it's done.

What AEO targets: featured snippets · People Also Ask boxes · voice search responses · direct-answer extraction inside AI chatbots

Content structure

Explicit Q & A sections — one question, one direct answer, then elaboration
Target question-format queries — "what is", "how to", "best way to"
Answer in the first 40–60 words of a section

On-page signals

FAQ schema — engines parse Q & A programmatically
HowTo schema — for process-driven content

Formatting

Short, declarative answer sentences
Use the exact question as the H2 / H3 heading
Response first, context follows

Pages that win snippets answer directly within the first paragraph, use the question verbatim as a heading, and have strong overall page authority. AEO isn't a separate content type — it's a structural discipline applied to existing content.

Exercise · 01 – 03

Good AEO or bad AEO?

Which version would a parser extract? Which one hedges? Which one commits?

Example 01 · Definition
"What is a landing page?"

Option A

"Landing pages have been used by marketers for many years as a way to capture leads and drive conversions. There are many factors that contribute to a landing page's effectiveness, and understanding these can help businesses achieve better results from their campaigns."

Option B

"A landing page is a standalone web page designed to convert visitors into leads or customers through a single focused call to action. Unlike a homepage, it removes navigation and distractions to keep the visitor focused on one goal."

Example 02 · How-to
"How do I improve my landing page conversion rate?"

Option A

"To improve landing page conversion rate: (1) reduce form fields to three or fewer, (2) match the headline to the ad that drove the click, (3) place your CTA above the fold, (4) add one specific testimonial with a measurable result."

Option B

"To improve your conversion rate, start by reviewing your analytics to understand where visitors are dropping off. You might also want to consider testing different headlines, adjusting form length, and ensuring your page loads quickly on mobile."

Example 03 · Comparison
"A/B testing vs multivariate testing — which is better?"

Option A

"A/B testing compares two versions of a page by changing one element at a time — best for lower traffic sites. Multivariate testing changes multiple elements simultaneously — requires significantly higher traffic to reach statistical significance."

Option B

"Choosing between A/B testing and multivariate testing depends on a number of factors including your traffic volume, the complexity of the changes, and how quickly you need results. Both approaches have their merits."

Schema

Schema markup — telling machines what content means.

Structured data added to HTML using the schema.org vocabulary — maintained by Google, Microsoft, Yahoo, and Yandex.

Without schema, the engine reads and guesses. With schema, you tell it directly — this is a product, this is a review, this is an FAQ.

Why it matters for AI search: systems extract prices, ratings, authors, Q & A as structured data — no prose interpretation needed. Schema-marked content surfaces more often in featured snippets, AI Overviews, and rich results.

Most useful types

FAQPage Article BlogPosting Organization Person Product BreadcrumbList HowTo

Tools: schema.org · Google Rich Results Test · Google Search Console

FAQ schema · example

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [{
    "@type": "Question",
    "name": "What is a landing page builder?",
    "acceptedAnswer": {
      "@type": "Answer",
      "text": "A landing page builder is a tool that lets marketing teams create standalone web pages without writing code."
    }
  }]
}

Drop this on a page with Q & A content and you become eligible for rich SERP listings — questions and answers expand right under your page title.

GEO

Generative Engine Optimisation — how it's done.

What GEO targets: being selected as a source when an AI synthesises an answer from multiple pages. The model doesn't return your page — it paraphrases, quotes, or cites it inside a generated response.

Pillar 01 · Citability

Specific, verifiable claims — numbers, dates, named sources
Avoid hedged language — "can help", "may improve", "research suggests" signal low confidence
Standalone sentences — a passage pulled from context should still make complete sense

Pillar 02 · Information gain

Contribute what the top 10 results don't already cover
Original data, proprietary research, firsthand case studies, named expert quotes
Generic summaries lose to the original source

Pillar 03 · Source signals

Named authors with verifiable credentials and a consistent publication history
Brand mentions & citations across third-party platforms — reviews, press, forums
Consistent entity information across the web — name, description, location, founding date

Pillar 04 · Content structure

Headings that mirror how questions are asked
Front-load the key point — retrieval scores passages, not pages
Single-idea paragraphs — easier to extract as a discrete passage

Citation Reality

Why AI won't cite you.

43%

of topically relevant pages — pages that contain the correct answer — receive zero citations from AI systems.

Getting cited is binary. Either you're in, or you don't exist.

The traffic math

Traditional CTR

15%

Standard organic clickthrough.

With AI summary

CTR collapses.

Conversion lift

23×

That 1% converts at 23× normal organic.

The funnel shrinks. The quality of what gets through is dramatically higher.

Failure Modes

Four reasons pages fail to get cited.

10.1%

Technical integrity

62.2%

Semantic alignment

27.1%

Content quality

0.6%

Systemic

01 · 10.1%

Technical integrity

Crawler can't access or parse the page.

FCC firewall served a CAPTCHA → bot moved on. JavaScript-rendered content invisible to many AI parsers. IMDb Jekyll answer buried under boilerplate.

02 · 62.2%

Semantic alignment

Readable, but misaligned with the query.

Turn of the Screw — query was about the 1898 novella, page was about the 1974 TV movie. Karnataka folklore sports buried under cricket and football.

03 · 27.1%

Content quality

Answer exists, but not extractable.

Doug Pederson — page had complete career stats in a dense table. No sentence said "he was a quarterback." AI cited a simpler page that stated it plainly.

04 · 0.6%

Systemic exclusion

Technically perfect, structurally outranked.

Small university ML course page was perfectly optimised. AI cited Coursera & edX instead. Training weights favour mega-platforms.

The biggest bucket is content, not infrastructure. So this is fixable.

GEO Tactics

What moves the needle.

Five tactics, tested across 10,000 queries.

1

Statistics addition — replace vague qualitative claims with hard numbers. "The Swiss eat 11 kg of chocolate per year" outperforms "the Swiss love chocolate." The model recognises the structural shape of credible data.
2

Quotation addition — embed direct quotes from named, reliable sources. Attributed human voices signal credibility.
3

Cite your sources — link to academic or authoritative references inside your content. The model was trained on a web where credible information comes with citations.
4

Fluency optimisation — kill passive voice, fix sentence structure, improve readability. Combined with stats addition, this produced a +32.1% visibility lift in testing.
5

BLUF — Bottom Line Up Front — place the core fact in the first sentence of each section. Parsers grab the top of each chunk first.

Agent GEO got a 40% improvement in citation rates by modifying only 5% of page content. Generic rewrites that altered 25% performed worse. Fix the plumbing — don't rebuild the house.

E-E-A-T is a Google quality framework standing for Experience, Expertise, Authoritativeness, and Trustworthiness.

Measurement

Being honest about what we can track.

Google Search Console

Query shifts. AI Overview inclusion on branded & category terms.

GA4 referral traffic

Sessions from perplexity.ai, chatgpt.com, copilot.microsoft.com, gemini.google.com. Incomplete, directional.

Manual prompt tracking

Run the same queries weekly across ChatGPT, Perplexity, Claude. Log what cites, what doesn't.

Emerging tools

Profound, Otterly.ai. Useful starting points. Treat as directional — none have privileged LLM access.

Suggested tracking sheet · 20–30 queries / week

Query	Platform	Brand mentioned?	URL cited?	Competitor cited?	Source type	Notes
best landing page builder for SaaS	Perplexity	No	—	Unbounce, Instapage	Review site	…

This Week · 1/2

Action checklist — technical & content.

Technical

Optimise basic SEO first
Check robots.txt — are GPTBot, ClaudeBot, PerplexityBot allowed?
Check sitemap.xml — all key pages, https, dates current
Audit one key page for JavaScript rendering issues — can a basic crawler read it without a browser?

Content

Pick your three highest-traffic pages — do they answer a specific question in the first paragraph?
Apply GEO best practices (citability, info gain, source signals)
Add an FAQ section with question-format headings & direct answers
Replace one vague claim per page with a specific number, named source, or cited statistic
Named author with a verifiable bio on every blog post

Do these in seven days and you're ahead of 90% of your competitors — most of whom still haven't audited their robots.txt.

This Week · 2/2

Action checklist — schema & entity.

Schema

FAQ schema on pages with Q & A content
Article or BlogPosting schema on all blog content
Organisation schema on your About page

Entity & ecosystem

Search your brand name on ChatGPT, Perplexity, and Claude — note the answer
Check presence on G2, Capterra, or relevant review platforms
Set up your manual prompt tracking sheet and run it once this week

Schema is one of the highest-leverage technical changes you can make. Takes hours. Pays back for years.

The Shift

In SEO, you optimise for the crawler and the ranker.

In AI search, you optimise for the retriever, the ranker, the summariser, and the citation layer.

Thank you. Questions?

rudranil@funnelysis.com

Search Optimisationfor AI Chatbots

Lets Analyze this Website

SEO still matters.

Two different optimisation games.

How AI chatbots actually answer.

The retrieval flow — six steps.

Who's visiting your site.

Let's analyse another website.

Two disciplines, one goal.

Answer Engine Optimisation — how it's done.

Good AEO or bad AEO?

Schema markup — telling machines what content means.

Generative Engine Optimisation — how it's done.

Why AI won't cite you.

Four reasons pages fail to get cited.

What moves the needle.

Being honest about what we can track.

Action checklist — technical & content.

Action checklist — schema & entity.

Search Optimisation
for AI Chatbots