“Ultimate guides” were the undisputed heavyweight champions of SEO. They were built specifically to align with how Google’s algorithm measured content value.
The “skyscraper technique” helped cement a doctrine: length = depth.
But the web moved on. Search intent shifted toward fast answers, AI saturation destroyed length as a credibility signal, and Google’s systems began penalizing the one thing ultimate guides were engineered to produce: zero information gain.
So, what now?
The new content constraint is extractability, and it changes every structural decision downstream, from brief to publication.
Your content has a word limit: the grounding budget
AI engines like Gemini allocate approximately 380 words per webpage for query grounding, regardless of the article’s total length. It’s a retrieval constraint you have to adapt to.
The extraction data is precise:
- Pages under 5,000 characters: 66% AI extraction rate.
- Pages over 20,000 characters: 12% AI extraction rate.
Generative systems now answer many queries without requiring a click. The traffic those pages once captured no longer exists to be captured. The 4,000-word ultimate guide content marketing approach actively destroys generative search visibility.
What replaces the informational library is something structurally different and considerably more demanding to produce. Every sentence must earn its place by naming an entity, stating a relationship, preserving a condition, or making a citable claim.
Dig deeper: How to write for AI search: A playbook for machine-readable content
See where your brand appears in AI search, where competitors are winning, and what it takes to become the answer AI recommends.
From keywords to positions: The padlock principle
Traditional keyword targeting asked one question: “What are people searching for?”
Problem-first positioning asks a harder one: “What situation has produced this search, and what does a genuinely useful answer look like inside that situation?”
That’s where the padlock principle becomes useful. Your business is a lock that opens for multiple combinations, each representing a distinct problem for a distinct person.
For example, a car insurance provider targeting “car insurance” is a category. The same provider building separate pages for “an 18-year-old new driver declined by standard insurers” and “a courier using a vehicle for commercial work” is a solution.
The distinction sounds philosophical until you realize it affects every downstream structural decision. Andrew Holland is right: AI killed low-grade informational SEO. Here’s some tactical advice to shift your content approach.
3 tactical rewrites for problem-first positioning
Replace categorical identity with problem identity
- Before: “We are an insurance provider.”
- After: “We solve the underwriting problem for first-time drivers under 25 who are declined by standard insurers.”
Rewrite titles as outcomes, not labels
- Before: “Car Insurance | BrandName”
- After: “Car insurance for new drivers under 25 declined by most providers”
Lean into constraints rather than suppressing them
Acknowledging that your solution works for teams of 100 or more but not for solo operators signals to a retrieval system that your content can be cited with confidence. Generic advice is the content AI already generates for free.
Constraint-aware, condition-specific guidance is what AI cannot replicate and therefore must source.
This logic collapses one of the most entrenched distinctions in digital marketing. The traditional separation between informational content and commercial landing pages was always somewhat artificial, but AI retrieval has made it structurally unsustainable.
What replaces the previous distinction is a fundamentally different content architecture: Every page is a document that knows exactly who it is for, states the problem it solves in the first sentence, and earns its keep by delivering a resolution specific enough to be cited but human enough to convert.
Marketers should start injecting problem-positioned, AI-readable answers directly into commercial pages rather than blogs. Low-grade information recaps like the “best tools for X” roundup and the “how-to” guide that adds nothing to existing knowledge have been absorbed by generative systems that now answer those queries without a click.
Dig deeper: How to keep your content fresh in the age of AI
Write for zero context
Every sentence must be self-contained and able to survive alone. AI retrieval systems do not read your article the way a human does: sequentially, with accumulating context.
Instead, an LLM will lift sentences in a “send this to someone without context” type of way by extracting passages and evaluating sentences as independent semantic units.
If a sentence requires its neighbors to make sense, it cannot be extracted and evaluated as an independent semantic unit (i.e., it’s neither easily understood nor useful for a machine).
The three failure patterns and their fixes:
| Failure | Example | Fix |
| Unresolved pronoun | “It also includes unlimited storage.” | “The Dropbox Business Standard Plan includes 5TB of encrypted cloud storage.” |
| Stripped condition | “The price has dropped significantly.” | “The Asana Enterprise Plan costs $24.99 per user per month, down from $30.49 in Q1 2024.” |
| Vague claim | “Our platform makes team management easier.” | “The Asana Enterprise Plan streamlines cross-functional project tracking for teams of over 100 people.” |
If you want to write LLM-friendly content, no matter what content format you are creating, here’s my advice: look into semantic triples.
Because AI systems evaluate content using identical retrieval infrastructure regardless of page type, the semantic triples (subject, predicate, object, conditions preserved) apply equally to blog articles, product descriptions, and pricing pages.
Here’s a concrete application of semantic triples: Make your heading more explicit. Explicit headings placed directly above their corresponding paragraphs add mathematical relevance (i.e., they improve cosine similarity scores), which means that an AI is 17.54% more likely to select that passage if it has a good headline.
The citation bait formula
How do you keep content fresh in the age of AI?
First, accept that you’re optimizing paragraphs, not pages.
The citation-bait formula defines how to structure the paragraph blocks that sentences belong to.
Step 1: Direct declarative opening (40 to 60 words)
No preamble. No “in this section we will explore.” The answer first, always. This block is what generative systems extract.
Step 2: Context (one to two sentences maximum)
Expand without burying. Every additional sentence beyond two reduces the density of what came before.
Step 3: Structured evidence
A table, a numbered list, or a comparison. Something extractable in its own right, independent of the surrounding prose.
Step 4: Self-contained heading
The H2 or H3 that follows must name the topic, intent, and scope of what just appeared. Not “Key takeaways.” Not “Overview.”
The heading must make complete sense when read entirely out of context, because in generative retrieval, it frequently will be.
The playbook for machine-readable content contains even more citation bait advice.
Adam Tanguay explains it very well: The authority layer compounds over time. This is why the citation bait formula works in both the short and long term.
Machine structure with human specificity
Managing the tension between AI-readable structure and human persuasion is difficult. Like Shrek’s onion analogy, LLM-friendly content has more layers than most people realize. You don’t have to choose between the two. You have to layer them.
The AI inverted pyramid places machine-readable answer blocks at the opening of each section. Human storytelling — the anecdote, the constraint, the actual number/stat/finding — belongs immediately after, connected by a natural transition that moves the reader from optimized structure and into earned narrative.
Jessica Foster identified Dove’s “Real Beauty Stories” as a great example of this type of copywriting. Dove opens with structured how-tos that satisfy intent-driven retrieval, then anchors those tutorials to the lived experiences of real customers.
The machine gets a citable answer at the top of the block. The human gets a reason to believe it in the body. Neither layer compromises the other because they occupy different positions in the document.
Casey Nifong has a great audit workflow for existing content:
- Identify the main question each section answers.
- Find the clearest direct answer buried in the paragraphs and move it to the top.
- Strip conversational lead-ins that delay the core answer.
- Run both the isolation test and the disambiguation test on every mid-page sentence.
- Leave stories, examples, and brand voice intact below the answer block, connected by natural transitions.
Track your visibility across AI search, uncover missed opportunities, and grow your presence where customers are asking questions.
The missing angle: Your workflow doesn’t exist yet
You now know good content no longer looks like a 4,000-word-long ultimate guide. Now it’s time to figure out what workflow produces said new good content.
Most articles on Search Engine Land describe the destination, not the road. That’s because you’re responsible for the journey. You need to build your editorial checklist, prompt structure (if you’re using LLMs to restructure existing content), and grounding budget calculation.
Go beyond theory and build an editorial system that consistently produces LLM-friendly content without sacrificing the human specificity no model can replicate.