Updated June 2026. This article was originally published in July 2025 as a synthesis of the best research available at the time. Twelve months and a lot of real audit work later, some of it has held up, some of it hasn't, and some of it deserves a sharper edge. This version reflects what we now believe based on doing this work for real clients across multiple verticals, not just reading other people's research. The original framing — that AI search is fundamentally different and your strategy has to adapt — was right. The specifics needed updating. Here's the honest take.
The search landscape has fundamentally shifted, and twelve months ago I wrote a piece arguing that AI-powered search was reshaping how content gets discovered. That part has only become more true. ChatGPT search, Claude, Perplexity, Gemini, and Google's AI Overviews are now the front door for a large and growing share of B2B buyer research — the latest data from G2's 2026 Answer Economy Report shows 51% of B2B software buyers now begin vendor research inside an AI chatbot rather than a traditional search engine, and Gartner is predicting a 25% drop in traditional search engine volume by 2026.
What changed is what we've learned about how to actually do something about it. The 2025 version of this piece read like a playbook — schema gives you a 30% lift, branded mentions are the #1 factor, smaller specialized sites can outrank major publications. All of those statements have a kernel of truth. None of them are the simple recipe they sound like. Here's what twelve months of running LLM visibility audits across pet insurance, virtual healthcare, e-commerce, professional services, and consumer brands has actually taught us.
It's still SEO. Best practices still apply.
The first thing worth saying clearly: this isn't a new discipline that replaces SEO. It's an extension of it.
The brands we audit that show up well in AI answers are, almost without exception, brands that do the boring fundamentals well. They have clear, well-structured content. They have schema markup on their key pages. They have a consistent technical foundation — clean URLs, fast pages, proper headings, internal linking that makes sense. They have content that's genuinely useful, written in the natural language of how their buyers actually talk about the problem.
If you've been doing real SEO work for the past decade, most of it still matters. Don't throw out your playbook. The mistake we see when consultancies pivot to "GEO" or "AEO" is treating it as a wholesale replacement — quietly shelving the SEO discipline and replacing it with vague directives about "optimizing for AI." That's a mistake. AI search rewards clarity and structure. Those are the same things classic SEO has always rewarded; the audience listening has just expanded.
LLMs are not the same. You cannot game them all.
The single biggest correction to the 2025 piece is this: I wrote about ChatGPT, Claude, Perplexity, and Gemini as if they were variants of the same thing, with different "personalities" but ultimately the same job. That framing undersells how different they actually are.
We've now audited the same brand across all four models in over a dozen engagements. What we see consistently is structural divergence, not just personality. A virtual healthcare company we audited had 94% accurate brand-mention rate on Claude and 55-57% on ChatGPT and Gemini. The brand hadn't done anything different between models — it's not that one model "knows" them and the others don't. The training data is different. The retrieval strategies are different. The way each model weights recency vs. authority vs. relevance is different. Some models pull from the live web with retrieval augmentation; others rely heavily on pre-training. Some prefer comprehensive answers; others prefer concise ones. Some cite sources transparently; others summarize without attribution.
The implication for strategy is significant: you cannot game four models simultaneously with one tactic. You cannot write a piece of content that ranks #1 across Claude, ChatGPT, Perplexity, and Gemini, because they are not running the same ranking system. The most you can do is increase your odds of being cited by giving each model the kinds of signals it tends to reward.
That sounds discouraging. It isn't. It means the goal isn't "rank in LLMs." It's "show up reliably enough, across enough models, that buyers researching your category find you." The threshold for winning is lower than ranking #1 on Google ever was — but the path is more about consistency across surfaces than dominance on any single one.
LLMs are your best visibility audit tool
Here's the framing that's become the most useful for us over the past year: the most valuable thing an LLM does for your brand isn't to rank you. It's to reveal you.
Ask Claude, ChatGPT, Perplexity, and Gemini the questions your buyers actually ask about your category, and what comes back is a free, real-time, machine-generated snapshot of how the internet has positioned you. It's a visibility audit you couldn't have run two years ago.
The gap between what you think you've communicated about your product and what the internet has actually communicated on your behalf is, in our experience, larger than most founders expect. Brands often discover that LLMs describe them in terms of their oldest, most-linked content — content that may not reflect how the company has evolved. They discover that competitors are getting cited in their answers because the comparison content out there was written by reviewers, not by them. They discover that the framing of their category — the words used to describe the problem they solve — has drifted from how they talk about themselves.
That gap is your roadmap. Not a ranking opportunity but a positioning opportunity. The LLM is telling you, plainly, what the internet currently believes about you. From there you can decide what to fix.
You influence LLMs by changing what the internet says
LLMs reflect the web. To change what they say about you, you have to change what the web says about you. There are two levers, and twelve months of work has convinced us both matter — but in different ways than the original 2025 piece suggested.
Lever one: your own content. This is where Mike's original piece was actually most right and also most under-stated. The single highest-leverage thing you can do is publish clear, structured content that explains, in your own words:
- What your product or service actually is, in plain language
- How it differs from alternatives in the market (named directly — comparison content is undercovered and over-rewarded)
- Where it sits in the competitive landscape — what category you're in, who you're for, who you're not for, the buying criteria a buyer in your category should be weighing
LLMs are pattern-matching machines. If you don't tell them how to position you, they'll position you the way other people have — typically more generically, often less flatteringly, and almost always with less precision than you'd choose.
We've found this matters more than schema, more than FAQ blocks, more than tactical formatting. The brands that show up well in LLM answers are the brands that have made it easy for the web to understand their differentiated value, not just find their pages.
Lever two: earned media and third-party citations. Here's where the original 2025 piece pointed in the right direction but the tactical advice was incomplete. Yes, branded mentions correlate strongly with AI citations (Ahrefs' 0.664 correlation number is still real). But the more useful insight a year in is this: LLMs are aggressive consumers of third-party content — reviews, comparisons, listicles, podcast transcripts, YouTube videos with timestamps and transcripts, Reddit threads, expert commentary in trade publications.
When we audit a brand and they're getting outpaced by a competitor, the gap is almost always traceable to third-party content. The competitor has more reviews on G2 or Capterra. They've been on more podcasts. They have more YouTube content (and Google now indexes YouTube transcripts heavily, which feeds Gemini). They show up in more "best of" listicles. They've been quoted in more trade press.
That's the playbook for reputation work in the LLM era. Run an audit, identify the third-party sources LLMs are citing for your competitors, and go earn presence in those same sources. Reddit, YouTube, podcast appearances, trade publication mentions, listicle inclusion — these matter more for AI visibility than they ever did for traditional SEO.
Freshness matters more than it used to
One thing the 2025 piece didn't emphasize that we now think is essential: freshness signals matter, a lot.
LLMs with retrieval augmentation increasingly favor recently published or recently updated content over evergreen pages that haven't been touched in years. A two-year-old "definitive guide to X" with strong backlinks can lose ground in LLM answers to a six-month-old article on a smaller site simply because the smaller piece is fresher.
This isn't unique to LLMs — Google has long used freshness as a signal for query-dependent topics. But LLMs amplify it. Our working hypothesis is that retrieval systems weight recency more heavily than classic ranking systems do, partly because they're trying to give answers that reflect the current state of the world, and partly because the underlying retrieval indices favor newer content for ranking purposes.
The practical implication is a refresh discipline. Pick your most important content. Update it on a schedule — quarterly is enough for most pages, monthly for fast-moving topics. Update the dateModified schema field. Rotate out stale stats. Add new sections as the topic evolves. We've seen multiple cases where a refresh on an existing page outperformed publishing an entirely new piece, by a margin that should not have been close.
Schema seems to help, but it's not a magic bullet
The 2025 piece cited a study showing pages with schema markup had an 8.6/10 accuracy score vs. 6.6/10 without — a 30% improvement in AI understanding. That number was real and the study was sound, but I oversold what it means.
Here's what we now believe, more carefully: schema markup is a best practice that helps LLMs understand your content more reliably, but it does not guarantee citation or visibility. We've audited brands with comprehensive schema implementations that don't show up well in AI answers, and brands with thin schema that do. The variance is high enough that schema alone is not a reliable predictor.
The same is true of FAQ blocks, H2/H3 hierarchy, bullet points, and other structural tactics that get a lot of airtime in "GEO optimization" content. They probably help. They're cheap to do. They're best practices for human readers anyway. Implement them. But don't expect any single one of them to be the lever that moves you from invisible to visible in AI answers.
The honest version: structural and schema tactics are necessary but not sufficient. They make your content easier for LLMs to parse, which gives you a better chance of being included in an answer when your content is the right content to include. They don't make your content into the right content to include. Only the substance of what you've written — clear, differentiated, educational, recent — does that.
What we'd revise from the 2025 version
A few specific updates to claims made in the original piece, in the spirit of intellectual honesty:
The "30% schema lift" number. Still likely true on accuracy of AI extraction. Not a ranking or visibility promise. Schema makes you legible, not preferred.
The "branded mentions are the #1 factor" framing. The Ahrefs 0.664 correlation is still real and still important, but correlation isn't causation. We've now seen brands with relatively low branded mention volume still surface well when their content is unusually clear and educational. The takeaway is that branded mentions matter for credibility, but they don't override poor content positioning.
The "smaller specialized sites can outrank major publications" claim. True in principle. Harder to execute than it sounded. The cases where this actually plays out are when the smaller site has done a much better job of educating LLMs on its specific niche than the larger publication has done generically. Specialization is a moat, but only if the specialized content is unambiguously better-structured and more current than the generalist competitor's.
The llms.txt question. A year ago I said evidence for llms.txt effectiveness was limited. That's still true. No major AI platform officially supports it as of mid-2026. Add the file because it costs nothing, but don't expect it to move anything.
What we'd add
A few things we'd add to the 2025 framing now that we've done the work:
Reputation management via LLMs is real and underserved. The audit-as-input workflow — query an LLM, identify what it's saying about you, identify the third-party sources it's citing, target those sources for earned media — is the most productive use of LLM visibility data we've found. It works because it converts the LLM from a ranking problem into a research tool.
Run real audits, not vibe checks. Most "I asked ChatGPT about my brand and here's what it said" exercises produce noisy, non-repeatable results. A real LLM visibility audit uses 80-100 prompts across all four major models, run in a clean-room context (no chat history, no system prompts), with consistent scoring. That data is dramatically more useful than ad-hoc queries.
The audit-as-input rhythm. Brands serious about LLM visibility should be running an audit quarterly. The space is moving fast enough that a snapshot from six months ago is already stale. We've shifted internally toward treating the visibility audit as a recurring instrument, not a one-time deliverable.
The bottom line, twelve months later
The 2025 version of this piece closed with the line "the rules of the game have fundamentally changed." That's still true. What we'd add now is that the rules have changed less radically than the early AI search discourse suggested. The fundamentals of clear, differentiated, current content with a strong technical foundation matter more than ever — they've just become legible to a new class of readers, the LLMs, that are increasingly the front door to your category.
The opportunity in 2026 isn't to "win" LLM rankings — that frame is a category error. The opportunity is to use LLMs as a brand visibility instrument that tells you, plainly, what the internet currently understands about you, and then to go fix the gaps in the source material. That's a strategic discipline, not a tactical checklist.
If you'd like to see what your brand looks like across Claude, ChatGPT, Perplexity, and Gemini today, that's exactly what we do. You can take a free Lite audit at loupeandblade.com/llm-visibility-audit, or get in touch with me directly at michael@loupeandblade.com.
Sources and further reading: research from Ahrefs (75,000-brand analysis), Kevin Indig's correlation research, Seer Interactive's SERP factor analysis, structured data testing by Aiso, the reported Claude algorithm analysis from GPT Insights, the G2 2026 Answer Economy Report, and Gartner search-volume forecasts. The AI search landscape continues evolving rapidly — this article will be revised again as the data warrants.
Frequently asked questions
Is GEO replacing SEO?
No. GEO (generative engine optimization) is an extension of SEO, not a replacement. The fundamentals of clear content, strong technical foundation, schema markup, internal linking, and brand-mention strategy that have always mattered for SEO continue to matter for visibility in AI answers. What's new is the audience reading your content — large language models are increasingly the intermediary between your content and your buyers — and the surface where buyers find you. The discipline is bigger, not different.
Can I "rank #1" in ChatGPT or Claude?
Not in the way you can rank #1 on Google. Each LLM has its own training data, retrieval strategy, and citation behavior. The realistic goal is to show up reliably across multiple models so that a buyer researching your category finds you regardless of which model they're using. Treat LLM visibility as a portfolio measurement, not a single-model ranking.
Does schema markup actually help with AI visibility?
It helps, but it isn't a magic bullet. Schema markup makes your content more legible to LLMs and improves the accuracy with which they can extract information about your business. But schema alone doesn't determine whether you'll be cited — the substance of your content (clarity, differentiation, freshness) matters more. Implement schema because it's a best practice; don't expect it alone to move you from invisible to visible.
What's the single highest-leverage thing I can do for AI search visibility?
Publish clear, differentiated content that explains what your product is, how it differs from alternatives, and where it sits in the competitive landscape — in language that matches how your buyers actually talk. Combine that with a refresh discipline (update key pages quarterly), schema markup on important pages, and a deliberate earned-media strategy targeting the third-party sources LLMs are citing for your competitors. None of these are exotic. All of them compound.
How do I know if my brand is showing up well in AI answers?
Run a real audit. Ask all four major models (Claude, ChatGPT, Perplexity, Gemini) the questions your buyers actually ask about your category. Score the answers for: are you mentioned, is the description accurate, are competitors mentioned, are competitors mentioned more positively. Do this across 80-100 prompts, not just a handful, and rerun quarterly. Anecdotal vibes from a single ChatGPT query are not data.
How fresh does my content need to be?
Fresher than you think. LLMs increasingly favor recently updated content, especially for fast-moving topics. We recommend a quarterly refresh on your most important pages — update stats, add new sections, update the dateModified field. For topics where the landscape moves monthly (AI, regulated industries, fast-moving consumer trends), refresh more often.
Do third-party sources like Reddit, YouTube, and podcasts really matter for AI visibility?
Yes — more than they ever did for traditional SEO. LLMs are aggressive consumers of third-party content because that's where independent, current, opinionated commentary lives. Reddit threads, YouTube videos with transcripts, podcast appearances, trade-publication quotes, and listicle inclusions all feed the corpus that LLMs draw from when answering questions about your category. If a competitor is showing up better than you in AI answers, look at where they're getting cited from. That's your earned-media roadmap.
No Comments.