CMSEO2025 Steve Toth.pdf

v

SEO’s Next Chapter:
Why Truth Beats
Traffic in the Age
of AI Search
Steve Toth

“LLM optimization is not an opportunity. It's THE
opportunity. AI, and LLMs, rip the fabric of marketing
open and sew it back together in new ways. Learning
how to influence LLMs as they influence buyers is
mission critical.”

A Quote from my colleague Kevin Indig …

v

About Me
●

Creator of AI Notebook & SEO Notebook — 22,000+
subscribers combined.

●

Left FreshBooks in March 2020, turning my former employer
into a client and grew my solo consultancy past $1 million in
its first year.

●

CEO of Notebook Agency - 30 person team.

●

Built award-winning strategies that rank #1 for multiple,
high-intent keywords with 300K+ monthly searches.

●

Dedicated thousands of hours to mastering AI search strategy.

1/3

v

About Me
●

Drove 50 million+ clicks for FreshBooks.com, including a
page worth $50K in weekly traffic.

●

Founded SEO IRL — Canada’s largest SEO conference.

●

Clients: Purple Mattress, Spellbook, FreshBooks, Royal Bank of
Canada, Mejuri, Cirque du Soleil, and Greg & David Baszucki of
Roblox.

●

Taught Answer Engine Optimization course for CXL ⭐4.7/5

●

My goal: help you stay two years ahead of your competitors.

2/3

Optimizing For
AI Refinements

v

Deep Research
●

Deep Research acts like an agent,
taking time to perform research.

●

It assembles a comprehensive
report that can take upwards of 10
minutes depending on the model
used and complexity.

Scanning Our
Client’s Website…

v

Hey Look,
We Did Alright!
In this presentation, I’ll share practical
tips to optimize for Deep Research.

How did we do it?

v

Advanced GEO:
Refinement Synthesis
●

ChatGPT Deep Research asks up follow-up questions after the
initial query.

●

Optimizing for these refinements ensures your content
answers these likely subsequent questions, maintaining
visibility in the conversation.

1/2

v

Advanced GEO:
Refinement Synthesis
●

AI search is conversational.

●

Users ask follow-up questions.

●

Optimize for refinement synthesis.

●

Ensures you make it to the finish line for refinements relevant
to your ICPs.

2/2

v

Cool! Where Can We Find
These Refinements?
●

ChatGPT Deep Research spells it out for us!

v

Refinement
Questions

v

But Wait! There’s a Bot
for it: Refinement
Synthesizer
●

Analyze model-specific
refinements

●

Highlight common threads

●

Eliminate redundant prompts

Here are some
of the Questions
v
it generated for

“Employer of
Record Software”

And here are the
Top Refinements
v
in order of
Importance

Main Refinement Themes
for Employer of Record Software:
From refinements:

Aggregated themes across models:

What content can we improve on the
website to better meet refinement criteria?
E.g. Entered the site Deel.com

v

Before we wrap up
Deep Research
●

Refinements will vary each time you run deep research or any
AI-based search on Google, Perplexity, Gemini etc.

●

It’s much better to identify the common refinement themes
and optimize pages towards that.

●

Target content on the passage level vs page level.

v

Common Refinement Themes
●

Comparisons: These help the LLMs reason more efficiently

●

Who it’s for? Aka, the ICP: These help the LLM personalize the
result

●

Reviews: Adds human authenticity

●

Price/budget: Goes without saying: important dealbreaker!

●

Integrations: Dealbreakers for your ICP

v

Wrapping up
Deep Research
●

Let ChatGPT guide your content.

●

Use the Deep Research Refinement Synthesizer to aggregate
refinements.

●

Optimize for refinement themes.

Making it to the AI
Journey Finish Line

So How Do We Increase Our
Chances of Attracting Our ICPs?
We optimize for Dealbreakers!

v

Optimizing for
Dealbreakers:
●

What causes customers not to choose your product or
service? Those are your "dealbreakers."

●

For example, let’s say a user wants a CRM that absolutely must
integrate with QuickBooks…

1/2

v

Optimizing for
Dealbreakers:
●

Dealbreakers are a common part of the LLM buyer journey.

●

LLMs ask users what their dealbreakers are to help find
solutions that best meet their needs.

2/2

Introducing Dealbreaker Detector

v

What’s a
Dealbreaker?
Deep Research surfaces dealbreaker
questions automatically…

v

What it Does:
●

Dealbreaker Detector is designed to help you identify all the
potential dealbreakers—those critical friction points—that
might cause a customer to walk away from a product or
service.

●

It simulates the mindset of a skeptical buyer and analyzes your
offering to flag issues across categories like:
○

Features (missing must-haves)

○

Integrations (compatibility gaps)

1/3

v

What it Does:
○

Fit: Limitations for users with specific needs

○

Pricing: Confusing or restrictive pricing models

○

Usability: Steep learning curves or poor UX

○

Compliance: Legal, privacy, or security concerns

○

Customer Support: Lack of help when needed

2/3

v

What it Does:
●

DD provides clear, categorized lists of specific customer
dealbreakers by thinking like a critical buyer, helping you
proactively fix objections before they cost you sales.

●

Identifies potential dealbreakers.

●

It also generates FAQ-style rebuttals to help you preemptively
address these objections in your marketing, sales, or
onboarding material.

3/3

v

Dealbreaker Questions

v

Optimizing for
Dealbreakers is THE way
to close deals inside the
black box of LLMs!

Creating LLM-Focused
Comparison Pages to
Dominate AI Search

v

Remember
this? →

v

The Power of LLM-Focused
Comparison Pages
●

Comparison pages, such as [Your Brand] vs. [Competitor], are
excellent assets for GEO. But to take it step further and truly
leverage the personalization that AI offers its users, we can go
even deeper with this strategy.

●

Two-way and three-way comparisons for ICP (Clickup vs.
Asana for Marketing Agencies)

●

Alternatives pages: (Asana Alternatives for Marketing
Agencies)

1/2

Comparison Page Example in Action
(found in ChatGPT)

Souce: https://www.zenpilot.com/blog/clickup-vs-monday.com

It’s working!

v

One Final Note on Comparison
and Alternatives Pages
●

Put dates in your title tags to trigger recency bias of AI search
prompts (use dynamic dates if your CMS supports it).

●

Recency bias is confirmed in many of these leaked AI system
prompts, e.g. Perplexity & ChatGPT

Creating an Explicit
AI Info Page

notebook.agency footer

v

What’s an AI Info Page?
●

An AI info page is a plain text page on your website you can
point AI’s to that contains important company information.

●

When visiting your homepage, the bot will follow the link and
process the information on the page and likely use it in their
response about your brand.

What’s an AI
Info Page?
Source

Info Pages in action…

Info Pages in action…

Info Pages in action…

Link to chat

v

Create your own
AI Info Page!
Link to the AI Info Page
Generator GPT

Optimizing for Truth
With Notebook Agency’s
Truth Alignment Framework™

The web is full of fake news

And so are LLMs

What happens when the answers
to your dealbreakers aren’t easily
retrievable, under your control,
dispersed, in accurate, incomplete,
or even misleading?

Fake news about your
brand likely exists

While “real news” aka the
Truth isn’t easily retrievable
and consolidated.

Let’s try a search…

Source

What caused Deel, Rippling
and Papaya Global to get
recommended?

And what about the qualified
companies that weren’t
recommended?

Like Oyster HR…

Source

So then why didn’t it appear
in the non-branded query’s
answer?

8 Different sources from
Oyster HR were needed to
get the answer and…

they were not easily retrievable…

In fact, Google highlights the
sources that it uses to generate
its AI answer!

Source

No wonder it didn’t make it in the
non-branded search!

Now, Deel on the other hand…

Source

Highlights = relevant passages
that support the query.

No wonder Deel made it into the
non-branded search!

Steps to interrogate:
●

Ask the non-branded
question…

●

And when you’re not
found, reformulate
and ask about your
brand and dig into
the sources!

So if we’re not making key
information retrievable, how
can we use Truth AlignmentTM
to boost visibility and
accuracy?

v

Actions You Can Take Today
●

Map your key dealbreakers on pages that the LLMs tend to
cite.

●

Optimize your help center documentation for LLM retrieval.

●

But to go deeper, we can apply the Truth Alignment
Framework™

v

Why Truth Alignment™ in
AI Conversations Matters
●

In AI Search decision-making, accuracy = opportunity.

●

AI is the new first impression.

●

These models aren't just surfacing brands. They’re making
recommendations.

●

The risk? Getting silently disqualified in black box AI
conversations.

1/2

v

Why Truth Alignment™ in
AI Conversations Matters
●

When your product truths like compliance, integrations, or
core capabilities aren’t consolidated on key pages and clearly
reflected in answers, LLMs may:
○
○

○

Recommend less capable competitors with
better-structured content
Misrepresent or oversimplify what you do by generating
an unfactual response based on inaccurate third-party
content.
Outright hallucinate when information is insufficient.

2/2

The Truth Alignment
Framework™ is Notebook
Agency’s effort to fix that… and
measure truth.

v

From the Starting Line to
the Podium
●

Visibility Isn’t the Win. Recommendation Is.

●

AI platforms are now the first and last touchpoints in many
buying journeys.

●

Being mentioned by LLMs isn’t enough, subsequent questions
cause LLMs to influence decisions.

●

Brands must be cited accurately, persuasively, and for the right
reasons.

v

What the Truth Alignment
Framework™ Does
●

Makes LLMs as knowledge as your top salesperson.

●

Turns LLMs into trained advocates, not just answer engines.

●

Ensures your brand is recommended (not just mentioned) in
late-stage queries.

●

Bridges the gap between brand truths and generative model
outputs.

v

Step 1: Codify Truth with
the Truth Notebook
●

Creating a single source of truth to measure against.

●

Centralized, client-validated record of product capabilities, ICP
fit, pricing, integrations, dealbreakers, and competitive
positioning.

●

Enables us to audit, score outputs against our ideal sales-grade
answers, and create citation-ready content.

v

Codify Truth with the
Truth Notebook™
●

We use ontology and taxonomy principles to create the Truth
Notebook (a sophisticated RAG system designed by our CTO
Max Geraci).

●

V1 is generated based on help documentation and information
like sales battlecards, internal documentation, sales call
transcripts, and even transcribed text provided by the client.

●

The Truth Notebook evolves as the product evolves.

v

Step 2: Interrogating LLMs
to find Truth Gaps
●

Use buyer-style prompts to test LLMs: “Does Recurly include
advanced subscription management, churn analytics, and
global payment support for scaling SaaS businesses?”

●

Responses are captured, graded, and analyzed across
ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews
for how well they reflect what’s in the Truth Notebook.

1/2

v

Step 2: Interrogating LLMs
to find Truth Gaps
●

What are the quality of the sources that influence AI’s
answers? Are they easily retrievable or incomplete and
scattered?

●

Do they fully represent the truth, partially or not at all?

●

Varied, natural language prompts test against different
phrasing.

2/2

v

Step 2: Interrogating LLMs
to Understand What They
Currently Recommend
●

We can even interrogate to see which products the LLMs
recommend one vs. the other.

●

With questions like:
○ Who has better customer support, Zoura or Recurly?
○

How does Zoura handle multi-currency compared to Recurly?

1/2

v

Step 2: Interrogating LLMs
to Understand What They
Currently Recommend
●

Which billing platform is better for managing complex
subscription upgrades and downgrades, Zuora or Recurly?

●

This measures the effectiveness of the comparison page
strategy.

2/2

v

Step 3: Score Outputs with
the Truth Score™
●

●

Each answer is scored on:
○

Accuracy (Is it fully correct, partially, or not at all? Are the
sources consolidated and clear? If not, then you’re less
likely to be included in nuanced non-branded prompts.)

○

Coverage (Are all key truths present?)

○

Sentiment (Are we being recommended?)

Benchmarks which truths are missing, distorted, undervalued
and not backed up with retrieval-friendly content.

v

Step 4: Detect and Fix
Discrepancies
●

●

Find Out Why AI Gets It Wrong
○

On-site issues (truths are not consolidated, content is
vague, not optimized for retrieval)

○

Off-site distortions (third-party content preferred over
first-party, unfavorable reviews, outdated articles,
competitor-driven narratives, i.e. their comparison pages)

Map LLM responses to the sources influencing them (possible
because we use browser automation and not API).

v

Step 5: Optimize for
Retrieval & Influence
●

Make the Right Truths Easy to Surface.

●

Rebuild on-site content to consolidate and support key
truths

●

Edit and saturate off-page content with correct truths
(websites must be in the same sphere and adding recency
language to these truths helps them get chosen for retrieval).

1/2

v

Step 5: Optimize for
Retrieval & Influence
●

Launch LLM-ready comparison pages structured for citations
and retrievable product truths.

●

Ensure truths appear in the passages LLMs are most likely to
quote.

2/2

v

Step 6: Monitor, Adapt, Repeat
to improve Truth Score
●

LLMs Evolve. So Should Your Representation.

●

Continuous re-interrogation with varied buyer prompts.

●

Real-time Truth Score dashboard.

●

Feedback loop informs content, PR, Social Media, & SEO.

●

Reach out to third-party websites to correct mistruths.

v

So yeah,
tracking
mere
mentions…

v

Without
an end?

v

To understand your
AI Share of Voice?

Nope!

Fact: AI answers change.
Tracking mention frequency
is not enough and not a true
reflection of reality.

Instead, we need to
Propagate and Track Truth &
on sources that LLMs cite!

Truth.

v

The Outcome: Sales-Grade
AI Representation
●

Your brand is not just mentioned, but chosen!

●

Greater influence in LLM buyer journeys.

●

More accurate citations and recommendation frequency.

●

Strategic defensibility against louder, less capable competitors.

v

Results

Source

v

One more?

Source

v

Ok, fine. Here’s another!

Source

Thank You!

Let’s Connect

How can we help?

ainotebook.com
(AEO Strategy
Newsletter)

seonotebook.com
(SEO Strategy
Newsletter)

notebook.agency

(my AEO/SEO Agency)

seoirl.com
(my conference)

Linkedin:

linkedin.com/in/stevetothjr



本文档为站内渲染。原始文件本地路径：saas/source/seo-llm/raw-seo知识库-TXT整理-seo方法论-ppt-John喜欢和引用的那些海外演讲嘉宾PPTs-CMSEO2025-b42f6c.txt（仅本地保留，不入库不部署）