AI Visibility Research

Michał Suski’s LLM Visibility Playbook

The practical retrieval system behind modern AI citations, prompt visibility, and brand proof.

Intro coming soon.

This analysis is based on public materials, third-party recaps, Surfer documentation, and SurferSEO articles available at the time of review. It should be treated as a strategic interpretation, not as a private or proprietary disclosure from Michał Suski or Surfer. We put this together because Michal is speaking at the SEO Conference in Chiang Mai that I was lucky to attend a few months before Covid. This document will be a great preparatory study guide to get more out of the presentation. Michal of course has no idea that we published this.

1. Executive Summary

Michał Suski’s recent approach to LLM visibility appears to be built around one simple idea: AI systems prefer sources that are easy to retrieve, understand, verify, and quote.

This is not traditional search engine optimization with a new label. It is not only schema markup, FAQ blocks, or writing more blog posts. It is a retrieval problem. The brand that wins is often the brand with the clearest answer, the strongest source footprint, and the lowest extraction cost.

Suski’s newer thinking appears to extend the original SurferSEO logic into the AI search environment. Surfer was built around analyzing top-ranking pages, identifying patterns, and helping teams optimize content against what search engines already reward. In the LLM era, the comparable move is to analyze which pages are cited by AI systems, which brands appear in generated answers, and which sources keep returning across related prompts.

The most useful strategic concept is Information Retrieval Cost. In plain English, this means the effort required for an AI system to find the answer inside a page or source. If a page hides the answer under a long introduction, vague positioning, weak headings, or unsupported claims, it has a higher retrieval cost. If the answer is direct, well-labeled, entity-rich, and supported by proof, it becomes easier for a model to use.

[LLM visibility is not only about being published. It is about being easy to extract.]

The practical lesson for marketers is significant. AI visibility work needs to move beyond content production. It needs prompt mapping, source tracking, citation analysis, page-level repair, third-party proof, and ongoing monitoring. The agency opportunity is to package those steps into a repeatable visibility system.

2. Why This Matters

Buyers are no longer using only search results to make decisions. They are asking AI systems to compare options, explain risks, summarize markets, and recommend providers.

That changes the role of content. A service page does not only need to rank. It needs to be understood by systems that assemble answers from multiple sources. A brand does not only need a strong website. It needs corroboration across the sources AI systems already use.

The search result is becoming an answer environment

In classic SEO, the visible unit of competition was the search engine results page. A marketer could inspect rankings, featured snippets, local packs, organic results, and paid placements. The visibility question was direct: which page ranks, in which position, for which keyword?

In AI search, the visible unit of competition is often the generated answer. That answer may mention brands without linking to them. It may cite third-party sources instead of the brand’s own website. It may use review platforms, publisher articles, directories, YouTube transcripts, Reddit threads, or competitor pages as supporting material.

This creates a new visibility problem. A company can rank well in Google and still fail to appear inside AI-generated recommendations. It can have a strong website and still lose the answer to a competitor with better off-site proof. It can publish long content and still be ignored because the answer is buried too deeply.

The core shift is from ranking pages to earning inclusion in generated answers. That means visibility needs to be measured at the prompt, answer, citation, and source level.

The old content checklist is too narrow

Many AI optimization recommendations still sound like traditional SEO advice: add FAQs, use schema, answer questions, write more content, and publish topical articles. Those items can help, but they are incomplete. They do not explain why AI systems cite one source, ignore another, or mention a competitor in a recommendation-style answer.

Suski’s more useful contribution is the shift from surface-level formatting to retrieval architecture. The question is not only whether the page contains the right keywords. The better question is whether the page makes the right answer obvious enough for an AI system to retrieve with confidence.

[The best AI visibility work reduces uncertainty for both machines and buyers.]

This has major implications for agency work. Content teams need to inspect the entire proof environment around a brand. That includes owned content, third-party validation, review language, expert bios, entity consistency, citations, and the way competitors are being described by AI systems.

3. The Core Method

Suski’s practical LLM visibility method can be reduced to a simple sequence: track the prompts, inspect the sources, identify the recurring winners, and repair the brand’s proof architecture.

This is the same strategic discipline that made modern SEO operational. You do not optimize from opinion. You inspect what the system already surfaces, then decide where to compete, where to join, and where to build supporting proof.

Step 1: Track prompts instead of only keywords

Keywords still matter, but AI search introduces a more conversational unit of demand: the prompt. A buyer may not search for “license defense attorney Los Angeles.” They may ask, “What should I do if I received an accusation from the California nursing board?” or “Who is the best attorney for a nurse license investigation in California?”

Those prompts carry intent, urgency, risk, location, and decision context. They also produce different answer types. One prompt may produce a legal explanation. Another may produce a comparison. Another may cite a government page. Another may recommend specific providers.

A useful AI visibility audit begins by building a prompt map. The goal is to identify the questions buyers actually ask when they are uncertain, urgent, or close to choosing a provider.

PROMPT TYPES TO TRACK

Discovery prompts: “best provider for [specific need]”
Risk prompts: “what happens if I receive [notice or problem]”
Comparison prompts: “[brand] vs. [competitor]”
Local prompts: “[service] in [city or state]”
Cost prompts: “how much does [service] cost”
Process prompts: “what should I do after [trigger event]”
Trust prompts: “who is qualified to handle [specialized issue]”

Step 2: Inspect which sources AI systems cite

The next step is to record the sources used by AI systems. This includes ChatGPT, Perplexity, Gemini, Google AI Overviews, and Google AI Mode where available. The useful question is not only whether the client appears. The useful question is which sources influence the answer.

In many categories, the cited source will not be the brand’s own website. It may be a directory, a review platform, a publisher list, a niche association, a government page, a YouTube video, a competitor article, or a user-generated discussion. Each source type creates a different action plan.

[AI citation analysis turns invisible competition into visible work.]

Step 3: Find recurring source patterns

The most important sources are not always the ones that appear once. They are the ones that appear repeatedly across related prompts. These recurring URLs or domains can be treated as core sources. They reveal where the AI system appears to find stable support for a topic.

A recurring source might be a competitor’s guide, a government explanation, a trusted review page, a high-authority publisher, or a comparison article. The correct response depends on the source. Some sources should be outranked. Some should be updated through profile optimization. Some should become targets for inclusion, quotes, sponsorship, digital PR, or directory improvement.

SOURCE DECISION MAP

Source Type	What It Means	Repair Action
Client-owned page	The brand has a direct citation opportunity.	Improve extractability, proof, headings, and coverage.
Competitor page	The competitor owns the answer structure.	Build a stronger page and compare topical gaps.
Directory or review site	AI is relying on third-party validation.	Improve profiles, reviews, categories, and descriptions.
Publisher or association	The topic needs external authority.	Pursue mentions, expert quotes, or inclusion.
Forum or community source	The prompt has subjective or experience-driven intent.	Study language, pain points, and unanswered concerns.

4. Information Retrieval Cost

Information Retrieval Cost is the most useful lens in Suski’s recent AI visibility work. It explains why a shorter, clearer, better-labeled source may outperform a longer but less extractable page.

The idea is simple: AI systems are more likely to use content that makes the answer easy to locate and support. The more a model has to infer, summarize, or dig, the more friction the page creates.

What low retrieval cost looks like

A page with low retrieval cost answers the main question quickly. It uses headings that match real user intent. It includes specific entities, names, dates, locations, credentials, examples, and constraints. It avoids long introductions that delay the answer. It uses comparison tables, short definitions, and direct answer blocks where they help.

This does not mean every page should be thin. It means depth needs structure. A long page can work well if the answer is easy to locate, each section has a clear purpose, and supporting proof is connected to specific claims.

[Depth helps only when structure makes the depth usable.]

How to repair a page for lower retrieval cost

A retrieval repair pass should happen after the core content strategy is clear. The editor’s job is to make the page easier for a human reader and an AI system to parse. That starts with the first screen.

The opening should state the answer directly. For example, a law firm page should not begin with ten lines about dedication, commitment, and experience. It should explain who the page is for, what problem it solves, what jurisdiction or category applies, and what action the reader should take next.

PAGE REPAIR CHECKLIST

Put the direct answer in the first 100 words.
Use headings that match real buyer questions.
Add definitions that make sense out of context.
Include specific services, locations, credentials, and examples.
Remove filler introductions and generic claims.
Add comparison tables where choices are confusing.
Connect claims to reviews, case examples, credentials, or source links.
Use internal links to related proof pages and expert bios.
Clarify who the service is for and who it is not for.
Make timelines, costs, risks, and next steps explicit where appropriate.

A retrieval repair is not the same as a copy edit. A copy edit makes language cleaner. A retrieval repair makes the answer easier to identify, verify, and reuse.

5. Prompt Tracking

AI visibility cannot be measured with a single yes-or-no check. A brand may appear for one prompt, disappear for another, and be described inaccurately in a third.

This is why prompt tracking needs to become part of the SEO reporting stack. The scorecard should include visibility, position, citation source, sentiment, and factual accuracy.

The new measurement layer

A practical visibility report should track how a brand appears across a defined prompt set. For each prompt, record whether the brand is mentioned, where it appears in the answer, whether it is cited, what source supports the mention, and whether the description is accurate.

This matters because AI systems can create partial wins. A brand might be mentioned but not recommended. It might be cited but described with outdated information. It might appear below competitors. It might receive a neutral description while competitors receive stronger proof language.

METRICS TO RECORD

Brand mention rate
Average answer position
Citation frequency
Source type distribution
Competitor mention share
Sentiment or recommendation strength
Accuracy of brand description
Prompt-level wins and losses

Why model-specific tracking matters

Different AI systems can produce different answers. ChatGPT, Perplexity, Gemini, Google AI Overviews, and Google AI Mode may rely on different indexes, retrieval methods, source preferences, and interface behaviors. That means one test result should not be treated as the market.

A useful report separates model behavior. If Perplexity cites review platforms, Google AI Overviews cites publisher pages, and ChatGPT mentions competitors without sources, each channel requires a different repair path.

[Prompt tracking shows where visibility is missing. Source tracking shows why.]

6. Core Sources

A core source is a URL, domain, or platform that repeatedly supports AI-generated answers for a topic. These are the sources worth studying hardest.

Core sources matter because they reveal the proof layer behind the answer. If the same directory, publisher, review page, or competitor article appears across many prompts, it becomes part of the category’s AI visibility infrastructure.

How to identify core sources

Start with a prompt set of 30 to 60 buyer questions. Run each prompt across the AI systems that matter to the client. Then record every cited source and every visible brand mention. The goal is to find recurrence.

Recurrence is more important than one-off visibility. A single citation may be noise. A repeated citation across multiple prompts signals that the source may have strong topical coverage, clear answer structure, entity relevance, or platform trust.

WHAT TO INSPECT

Which URLs appear across multiple prompts
Which domains appear across multiple AI systems
Which competitors are repeatedly mentioned
Which third-party platforms influence recommendations
Which source types appear for commercial prompts
Which source types appear for informational prompts

How to act on core sources

Once the recurring sources are known, the team needs a response plan. If a competitor page is the core source, the task may be to build a more complete and more extractable page. If a directory is the source, the task may be profile optimization and review improvement. If a publisher list is the source, the task may be digital PR or expert inclusion.

The wrong move is to treat every missing citation as a blog content problem. Some visibility gaps are website problems. Others are reputation problems, authority problems, review problems, or source-distribution problems.

[The repair action depends on the source, not the symptom.]

7. Off-Site Proof

A brand’s website is only one part of its AI visibility footprint. Many AI systems rely on third-party sources when answering recommendation, comparison, and trust-based prompts.

This is where classic SEO, digital PR, reputation management, and content architecture begin to overlap. The brand needs proof that travels beyond its own domain.

The website is not enough

Owned content is essential, but it cannot carry every trust signal alone. A company can claim expertise on its service page. A stronger proof environment connects that claim to reviews, case examples, expert bios, directory profiles, publisher mentions, videos, and consistent third-party descriptions.

In AI answers, off-site sources can become the deciding layer. When the prompt asks for the “best,” “most trusted,” “top-rated,” or “recommended” provider, AI systems often need corroboration beyond the brand’s own website.

Off-site proof does not replace the website. It supports the website by giving AI systems and buyers more places to verify the same entity.

The proof footprint to build

A strong proof footprint should repeat the same facts across multiple credible surfaces. That includes services, locations, specialties, founder names, credentials, categories, reviews, examples, and differentiators. Consistency matters because AI systems assemble answers from fragments.

AI SOURCE FOOTPRINT

Official website pages
Founder and expert bios
Case studies or representative examples
Google Business Profile and review platforms
Industry directories and niche listings
Publisher mentions and expert quotes
YouTube videos with clear transcripts
Podcast appearances and summaries
Association profiles and credentials
Comparison pages and buying guides

[AI visibility improves when the same proof appears clearly across trusted surfaces.]

8. Agency Application

For an agency, Suski’s method points toward a clear service opportunity: sell AI visibility as a proof and retrieval system, not as a content gimmick.

This positioning is stronger than “we optimize for ChatGPT.” It explains the actual work. The agency identifies prompts, tracks answers, studies sources, repairs pages, strengthens off-site proof, and monitors movement over time.

Package 1: AI Visibility Audit

The audit is the entry point. It shows the client where they appear, where they are missing, which competitors are winning, and which sources shape the AI answer environment. This is especially valuable for law firms, medical practices, home services, SaaS companies, professional services, and local businesses with high-trust buying journeys.

AUDIT DELIVERABLES

Prompt map with 30 to 60 buyer prompts
Brand visibility score by model
Competitor mention analysis
Citation source inventory
Core source identification
Accuracy and sentiment review
Owned page retrieval audit
Off-site proof gap map

Package 2: AI Visibility Optimization

Optimization is the repair phase. This is where findings become editorial, technical, and distribution work. The agency rewrites service pages, improves answer blocks, adds proof, strengthens entity consistency, updates profiles, and builds third-party validation.

OPTIMIZATION WORK

Service page rewrites for lower retrieval cost
FAQ and answer-block creation
Expert bio expansion
Internal proof linking
Review profile improvements
Directory and listing cleanup
Comparison content creation
YouTube transcript strategy
Publisher and expert quote strategy

Package 3: AI Visibility Monitoring

Monitoring turns the work into a recurring service. The client needs to know whether competitors are gaining mentions, whether AI descriptions are accurate, whether citations have shifted, and whether new source opportunities have appeared.

[AI visibility is not a one-time content task.
It is a recurring measurement and repair system.]

9. Implementation Framework

The best way to apply this method is to start small, choose one niche, and build a repeatable operating system.

A 30-day pilot is enough to prove the model. The goal is not to fix every visibility gap immediately. The goal is to create a clear map of prompts, sources, competitors, and repair actions.

Phase 1: Build the prompt set

Start with one client category. For example, a California professional license defense firm could track prompts around accusation letters, licensing board investigations, license revocation, settlement options, and attorney selection. A dental implant practice could track prompts around cost, candidacy, risks, procedure timelines, and provider comparisons.

The prompt set should include early research, urgent problem, comparison, local, and decision-stage questions. Each prompt should represent a real buyer hesitation.

Phase 2: Run the visibility audit

Run each prompt across the AI systems that matter. Record the answer, brand mentions, competitor mentions, citations, source types, and accuracy problems. Do not rely only on one model. Do not rely only on API responses if the goal is to understand what buyers see in consumer-facing interfaces.

Phase 3: Create the repair map

Group the findings into action categories. A page problem needs a page repair. A directory problem needs profile work. A publisher problem needs inclusion strategy. A review problem needs reputation work. A missing proof problem needs evidence, examples, credentials, or internal linking.

Finding	Likely Cause	Repair Action
Client is not mentioned for urgent problem prompts.	Service page does not answer the trigger event directly.	Create or rewrite the page around the buyer’s urgent question.
Competitor is cited repeatedly.	Competitor owns a clearer guide or stronger source footprint.	Build a more complete page and support it with proof links.
Directory pages influence recommendations.	AI is relying on third-party validation.	Improve profiles, categories, reviews, and descriptions.
Brand appears but description is inaccurate.	Entity facts are inconsistent or outdated across sources.	Clean up website copy, bios, listings, and public profiles.

[A repair map must be concrete enough to assign work.]

10. Cautions

AI visibility is real, but it is easy to oversell. The strongest teams will avoid hype and build disciplined measurement systems instead.

The market will be crowded with shortcuts: fake authority, bulk content, schema promises, AI-generated pages, and dashboard-only reporting. Most of those tactics will miss the actual problem.

Do not reduce this to schema

Schema markup can help machines understand page elements, but it does not create expertise, trust, reviews, topical coverage, or third-party validation by itself. A weak page with schema is still a weak page.

Do not rely on content volume

Publishing more pages does not automatically improve AI visibility. If the pages repeat generic claims, fail to answer real prompts, and lack proof, they may increase noise without improving retrieval.

Do not ignore source diversity

A brand may need visibility on review platforms, directories, publisher pages, videos, association profiles, and comparison content. Treating the company website as the only source of truth limits the visibility strategy.

The anti-hype lesson is simple: AI visibility is not won by one trick. It is built through consistent proof, clear answers, and measurable source coverage.

11. Sources and Further Reading

The following sources informed this strategic interpretation of Michał Suski’s recent public LLM visibility methods.

PUBLIC SOURCES REVIEWED

The Practical Conclusion

Michał Suski’s recent LLM visibility work points to a disciplined, measurable, and highly practical model. The goal is not to chase AI systems with tricks. The goal is to make a brand easier to retrieve, understand, verify, quote, and recommend.

For agencies, this creates a clear service architecture: prompt tracking, citation analysis, retrieval repair, off-site proof building, and recurring AI visibility monitoring.

The winning brands will not simply publish more content. They will build better proof systems. Want to see where your brand stands in AI answers? Request a free site audit.