Why your technical docs may be your most valuable digital asset

Documentation answers the questions buyers actually ask, which makes it bottom-of-funnel content. What LLM traffic is worth, what retrieval rewards, and which pages to fix first.

GTM
August 18, 2026
·
6 min
read
Abstract brand form on a purple field

Your documentation is bottom-of-funnel content, and most companies still fund it like a cost center they'd close tomorrow if the support queue allowed it. It gets a contractor, a backlog, and apologies. Meanwhile the buyer three weeks into evaluating you is reading it at eleven at night, working out whether the product does the specific thing they need, and they'll never fill in a form to ask.

That was true long before generative engines existed, and what changed is that a second reader showed up, reading the same pages for the same reason in roughly the same way. Retrieval doesn't read a website the way a person reads a brochure, front to back and in the order you chose. It works more like someone pulling one part off a shelf in a workshop, where the labelled shelf wins and the unsorted pile loses, however good the parts in that pile happen to be. Documentation is the labelled shelf. Most of the rest of a B2B website is the pile in the corner.

So what's that traffic actually worth, and which pages do we fix first?

Is small traffic bad traffic?

Not when it converts. Patrick Stox published Ahrefs' own analytics for a 30-day window in June 2025, and AI search sent 0.5% of their total traffic while producing 12.1% of their signups, a conversion rate 23 times higher than traditional organic search. Twelve percent of signups from half a percent of visits!

Should we rebuild a website around half a percent of its traffic? Well, of course not, and Stox himself states two caveats that are worth reading before anything gets moved. Visitors from AI search click links about 75% less than they do in organic search, and he isn't sure the pattern scales, quoting a colleague who expects these click rates are the highest they'll ever be. So the multiple will come down. The direction holds anyway, and it matches practice: this traffic is small in volume, late in the cycle, and it lands after the comparison work has happened somewhere we can't watch.

Which is why the number to manage is qualified traffic rather than raw traffic, tagged with UTMs and judged on the leads it converts. This reader turns up at the end of their research rather than at the start of it. So the pages worth optimizing are the end-of-funnel ones: pricing, comparisons, integration limits, security posture, and the documentation that answers what a sales call would have answered.

What makes a document trustworthy is what makes it retrievable

An engineer scanning a reference page wants one section, wants to confirm it's current, and wants to leave. They skim headings. They trust a definition that defines, they distrust a page that buries the specific inside the general, and they abandon anything that reads like it was written to fill a content calendar. Retrieval behaves in a recognizably similar way, because it works with pieces of your content rather than the whole polished narrative, and the pieces it can use are the ones that stand on their own.

So why do the two disciplines keep issuing the same instructions? Because they're solving the same problem: a reader who needs a precise answer and has no patience for the paragraph it's buried in. No fluff, high accuracy, strong structure, lean content, clear hierarchy.

What a technical reader needsWhat retrieval needsThe move for both
To land on one sectionPassages that stand aloneH2s phrased as the question
Terms defined the same way everywhereOne canonical passage per conceptA definitions page everything links to
Proof the page is currentFreshness it can checkLast-updated dates, versioned pages
Values, not proseStructure it can liftTables for limits, tiers, specs, versions
The next answer one click awayLinks between related answersDeliberate internal linking

I've built that shape several times, and none of it was built for retrieval, because retrieval wasn't a consideration in most of those years. At Lightrun, a 250-page knowledge base from a blank slate in eight weeks. At Firebolt, 100+ pages standing before public launch, and at Lytx and Surfsight, a 300+ page developer portal. At Coro, a compliance hub covering more than 20 global regulatory frameworks, built as an always-on resource rather than a campaign. At Visual Layer I documented the entire product codebase, and it came back from the engineers who wrote that code with close to zero corrections. That's the only quality test for technical content that's ever meant much to me.

None of this is a new argument either. A deck I wrote in March 2021 had a slide titled "Doc use cases: the bottom line", and its first line read "leverage docs for additional SEO + POCs and decision-making". So were docs funded as a discovery channel back then? They weren't, and mostly they still aren't, even though they were already the layer that buyers evaluated us on.

So is llms.txt the fix?

Not on its own. A mapping file or an llms.txt is an internal planning tool that forces us to decide which page owns which question. That's genuinely useful, and it's mostly useful to us rather than to the machine, because the work that changes outcomes is duller.

  • FAQs written the way a person asks. Somebody types "does it work behind a proxy", not "network configuration considerations".
  • The follow-up question, answered underneath it. Every answer produces a second one.
  • Target terms inside the H2s. Structure carries as much weight as keyword presence.
  • Internal links between the answers. Cheap, unglamorous, and usually skipped.
  • Natural writing instead of stuffing. A relevant subject contains the right terms anyway.

One warning belongs in bold here: don't damage your Google rankings while you optimize for answer engines. SEO isn't over. It has been declared dead several times before, and every time the answer came back the same, which is that it needs continuous adjustment rather than a funeral.

How do we know any of it is working?

Partly through tooling, carefully: on a competitive engagement for a B2B SaaS client we mapped their visibility against five competitors and evaluated five AI visibility and brand-monitoring tools along the way. The tools disagree with each other, sometimes badly, which is what you'd expect from a category this young. So we treated their numbers as a baseline to move rather than a score to win, and put the weight on GA4 and UTM-tagged traffic, where conversions are countable and arguments are shorter.

The limit is worth stating plainly, because restructuring your documentation won't rescue positioning that was never clear in the first place. It won't make a model recommend you over a genuinely better product, and it can't repair a claim your buyers already doubt. What it does is make you findable, quotable, and easy for a skeptical reader to verify. Whether the reader likes what they verify is a separate piece of work, and it happens well upstream of anything in this post.

Key takeaways

  • Docs answer evaluation questions, which makes them bottom-of-funnel content.
  • Qualified traffic is the number that matters, so tag it with UTMs and count leads.
  • Technical writing and answer-engine optimization ask for the same things: accuracy, hierarchy, definitions, tables, links.
  • llms.txt is planning. FAQs, follow-up questions, headings, and internal linking are the work.

The Snyk post on PCI and open source security requirements was built this way in 2019, long before anyone said GEO, and it holds up because the structure did the work rather than the adjectives. If your documentation is accurate and nobody can find anything in it, that's fixable, and it's cheaper than the campaign you were about to run instead. Bring the sitemap to a free 30-minute call, and we'll pick the ten pages your buyers evaluate you on and decide which question each one owns.

zero-click-search
Zero-Click Search
A search where the user gets the answer on the results page and never clicks through to a website. AI summaries and featured snippets drive it.
Text Link
structured-data
Structured Data
Standardized markup, mainly JSON-LD using the Schema.org vocabulary, that tells search engines and AI systems what a page is about: its entities and their relationships.
Text Link
share-of-voice-ai
Share of Voice (AI)
The percentage of AI-generated answers that mention or recommend a brand, compared with its competitors, across a fixed set of queries.
Text Link
perplexity-ai
Perplexity AI
An AI answer engine that pairs language models with live web search. It returns conversational answers with visible source citations and favors recent content.
Text Link
llm-optimization
LLM Optimization
Optimizing content to improve visibility, citation and favorable representation in answers from large language models such as ChatGPT, Gemini and Perplexity.
Text Link
knowledge-graph-optimization
Knowledge Graph Optimization
Building and maintaining a brand's presence in knowledge graphs, which are structured databases of entities and their relationships, to improve visibility in AI answers and knowledge panels.
Text Link
google-ai-overviews
Google AI Overviews
AI-written summary boxes at the top of Google results. They combine information from several sources to answer the query directly.
Text Link
google-ai-mode
Google AI Mode
Google's conversational search interface. It gives longer, multi-source answers with many citations and lives beside the standard results page.
Text Link
generative-engine-optimization-geo
Generative Engine Optimization (GEO)
Optimizing content to improve a brand's visibility and citation rate in AI search engines such as ChatGPT, Perplexity and Google AI Overviews.
Text Link
e-e-a-t
E-E-A-T
Google's quality framework: Experience, Expertise, Authoritativeness and Trustworthiness. It guides how Google evaluates content, and it matters for AI citation too.
Text Link
citation-rate
Citation Rate
How often a brand, site or page is referenced and linked in AI-generated answers for a given set of prompts.
Text Link
chatgpt-search
ChatGPT Search
OpenAI's search feature inside ChatGPT. It pairs a language model with live web search and answers with inline source citations.
Text Link
barnacle-geo
Barnacle GEO
A Generative Engine Optimization tactic: earn mentions on high-authority third-party pages that AI engines already cite, instead of relying only on your own site.
Text Link
answer-engine-optimization-aeo
Answer Engine Optimization (AEO)
Optimizing content to be chosen as the direct answer in AI search engines, voice assistants and featured snippets.
Text Link
ai-friendly-content
AI-Friendly Content
Content written and structured so AI search engines can parse it, extract answers from it and cite it.
Text Link
ai-visibility-score
AI Visibility Score
A composite metric that rolls a brand's mentions, citations, sentiment and position across AI search engines into one benchmark number.
Text Link
ai-search-ranking
AI Search Ranking
The position, prominence and frequency with which a brand or source appears in AI-generated answers. It follows different factors than classic search rankings.
Text Link
ai-citation
AI Citation
A source link an AI search engine attaches to its answer, crediting a specific page, domain or brand for the information.
Text Link
ai-brand-mention
AI Brand Mention
An instance where an AI search engine names a brand, product or company in its generated answer, with or without a source link.
Text Link
ai-visibility
AI Visibility
How often and how favorably a brand appears in answers from AI search engines such as ChatGPT, Perplexity and Google AI Overviews.
Text Link
ai-search
AI Search
Search that uses large language models to write a synthesized answer instead of returning a ranked list of links. ChatGPT Search, Perplexity, Google AI Overviews and Gemini are examples.
Text Link
risc-relevance-impact-strategic-alignment-cost
RISC (Relevance, Impact, Strategic Alignment, Cost)
A weighted scorecard for deciding whether an event earns a booth, a badge, or nothing. RISC stands for Relevance, Impact, Strategic alignment, and Cost. Weights are set before the sponsorship prospectus arrives.
Text Link
share-of-voice
Share of Voice
The portion of market conversation a brand owns compared with its competitors: mentions, search visibility, and coverage. It is a read on how much of the category talks about you.
Text Link
feature-activation-rate
Feature Activation Rate
The share of eligible users who use a new feature meaningfully after it launches. It shows whether a launch changed customer behavior, not just whether it created awareness.
Text Link
annual-contract-value-acv
Annual Contract Value (ACV)
The average revenue a customer contract brings in per year. It is used to size deals, compare segments, and track how much new business a company signs in a period.
Text Link
deal-velocity
Deal Velocity
How quickly deals move through the sales pipeline from first contact to close. Shorter sales cycles mean less friction between a buyer's interest and a decision.
Text Link
win-loss-ratio
Win/Loss Ratio
The share of competitive sales opportunities a company wins compared with those it loses, tracked over a period. A rising ratio suggests that positioning and enablement are working.
Text Link
battlecard
Battlecard
A short internal reference that helps a sales rep win against one named competitor: how they position, where they are strong, where they are weak, and how to answer the objections buyers raise.
Text Link
sales-enablement
Sales Enablement
The training, content, and tools that help sales teams sell: product education, talk tracks, decks, and competitive material, kept current and shaped by feedback from the field.
Text Link
product-market-fit
Product-Market Fit
The point at which a product satisfies a clearly defined market: the right customers adopt it, keep using it, and tell others. The first target for an early-stage company.
Text Link
go-to-market-gtm
Go-to-Market (GTM)
The plan and ongoing practice for bringing a product to its buyers: audience, positioning, pricing, channels, launch, and feedback. It runs as a cycle that continues after launch day.
Text Link
ideal-customer-profile-icp
Ideal Customer Profile (ICP)
A description of the type of company most likely to buy, succeed with, and stay with a product. It combines firmographics, technology stack, challenges, and budget reality into a portrait precise enough to act on.
Text Link
messaging-framework
Messaging Framework
A single document that turns positioning into the words a company uses: the core narrative, the value propositions, and the proof points for each audience. Product, marketing and sales all draw from it.
Text Link
positioning
Positioning
The place a product holds in the buyer's mind relative to the alternatives: who it is for, what problem it solves, and why it is different. It is the foundation every message is built on.
Text Link
data-augmentation
Data Augmentation
Creating extra training examples by altering existing ones, such as flipping, cropping, rotating or recolouring images, so the model sees more variety.
Text Link
workflow
Workflow
The ordered steps and hand-offs by which work moves from start to finish, who does each one and what triggers the next.
Text Link
verification
Verification
Confirming that a result, label or component is correct by checking it against a reference or a second method.
Text Link
vectorized-execution
Vectorized Execution
A query-engine technique that processes whole columns of values in one CPU instruction instead of row by row, which is what makes engines like DuckDB fast.
Text Link
vector-space
Vector Space
The multi-dimensional space where embeddings live; distance within it stands for how similar two items are.
Text Link
uncertainty-sampling
Uncertainty Sampling
An active-learning strategy that sends the examples the model is least sure about to human reviewers first, because those teach it the most.
Text Link
transformer
Transformer
A neural network architecture built on attention, which weighs every part of the input against every other part; it underpins modern language and vision models.
Text Link
traceability
Traceability
The ability to follow any output back through the data, labels, code and model versions that produced it.
Text Link
tokenization
Tokenization
Splitting text into the units a language model works with, which may be words, parts of words or characters.
Text Link
token-limit
Token Limit
The maximum number of tokens a language model can take in and produce in one request, which caps how much context it can hold.
Text Link
threading
Threading
Running several parts of a program at the same time on separate threads so work overlaps instead of waiting in line.
Text Link
temporal-analysis
Temporal Analysis
Studying how things change over time across frames or records, to detect motion, events, trends or anomalies.
Text Link
sla
SLA
Service Level Agreement: the committed level of service a provider promises, such as uptime, response time or support hours, with consequences if missed.
Text Link
simulation
Simulation
Generating realistic scenarios in software to test or train a system where real data is scarce, risky or expensive to collect.
Text Link
similarity-search-vector-search
Similarity Search (Vector Search)
Finding the items whose embeddings sit closest to a query's embedding, which returns things that look or mean alike rather than share keywords.
Text Link
setup
Setup
The initial configuration of a project: connecting data sources, choosing models and settings, and defining the workflow before work begins.
Text Link
semi-supervised-learning
Semi-Supervised Learning
Training on a small labelled set plus a much larger unlabelled one, using the labelled examples to guide what the model learns from the rest.
Text Link
semantics
Semantics
The meaning behind data rather than its surface form, which lets a system treat "car" and "automobile", or two photos of one scene, as the same thing.
Text Link
segmentation
Segmentation
Dividing an image into regions by labelling every pixel, so the exact shape of each object or area is known.
Text Link
search-attributes
Search Attributes
The fields a search can filter or rank on, such as labels, tags, dates, camera id or model confidence.
Text Link
search-and-retrieval
Search & Retrieval
Finding the items in a collection that best match a query, whether by keyword, metadata or visual and semantic similarity.
Text Link
scene-understanding
Scene Understanding
Interpreting a whole image or video as a scene: what objects are present, where they are and how they relate.
Text Link
saliency-map
Saliency Map
A heat map over an image showing which pixels most influenced a model's prediction, used to explain and debug it.
Text Link
reward-function
Reward Function
In reinforcement learning, the rule that scores each action so the agent learns which behaviour to repeat.
Text Link
review-batch
Review Batch
A set of items grouped for a human reviewer to check in one sitting, often chosen because the model was unsure about them.
Text Link
relational-database
Relational Database
A database that stores data in tables with rows and columns and links them through keys, queried with SQL.
Text Link
regularization
Regularization
Techniques that stop a model memorising its training data, such as penalising large weights or dropping random neurons, so it generalises better.
Text Link
refactoring
Refactoring
Restructuring existing code to make it cleaner and easier to change without altering what it does.
Text Link
recipe
Recipe
A saved, repeatable set of processing steps or settings that can be rerun on new data to get the same kind of result.
Text Link
rag
RAG
Retrieval-Augmented Generation: a language model that first fetches relevant documents and then writes its answer from them, which cuts hallucination.
Text Link
quality-assurance-qa
Quality Assurance (QA)
The checks and processes that make sure a product, dataset or model meets its standard before it ships.
Text Link
predictive-maintenance
Predictive Maintenance
Using sensor data and models to forecast when equipment will fail, so it is serviced just before it does rather than on a fixed schedule or after a breakdown.
Text Link
policy
Policy
In reinforcement learning, the strategy that maps each state to the action the agent should take.
Text Link
pixel
Pixel
The smallest unit of a digital image, a single point with a colour or intensity value; images are grids of them.
Text Link
pgvector
pgvector
A PostgreSQL extension that stores embeddings and runs similarity search inside the database, so vector search needs no separate system.
Text Link
one-shot-learning
One-shot Learning
Learning to recognise a new class from a single example, usually by comparing against a learned notion of similarity.
Text Link
object-storage
Object Storage
Storage that keeps files as objects with metadata in flat buckets rather than folders, built for huge volumes and cheap durability.
Text Link
normalization
Normalization
Rescaling values to a common range or distribution so no single feature dominates and models train more stably.
Text Link
noise
Noise
Random or irrelevant variation in data that hides the real signal, from sensor error, compression, lighting or sloppy labels.
Text Link
mvcc
MVCC
Multi-Version Concurrency Control: a database technique that keeps multiple versions of a row so readers and writers do not block each other.
Text Link
monitoring
Monitoring
Continuously watching a deployed system's health and output quality so drift, errors and slowdowns are caught early.
Text Link
model-versioning
Model Versioning
Tracking each trained model with its data, code and settings, so any version can be reproduced, compared or rolled back.
Text Link
metrics
Metrics
The numbers used to judge a model or system: accuracy, precision, recall, F1, latency and others chosen to match the goal.
Text Link
metric-learning
Metric Learning
Training a model so that similar items end up close together in its embedding space and dissimilar ones far apart.
Text Link
meta-learning
Meta-learning
Training a model to learn how to learn, so it adapts to a new task from very few examples.
Text Link
mask
Mask
A pixel-level map marking exactly which pixels belong to an object or region, the output of segmentation.
Text Link
markov-decision-process-mdp
Markov Decision Process (MDP)
The mathematical frame for reinforcement learning: states, actions, rewards and transition probabilities where the next state depends only on the current one.
Text Link
loss-function
Loss Function
The formula that scores how wrong a model's predictions are during training; training works to make this number as small as possible.
Text Link
latent-feature
Latent Feature
A hidden characteristic a model learns internally that is not directly observed in the data but helps explain it.
Text Link
label-noise
Label Noise
Errors and inconsistencies in labels, from mistakes, ambiguity or disagreeing annotators, which degrade what a model can learn.
Text Link
keypoint-detection
Keypoint Detection
Locating specific points of interest on an object, such as joints on a body or corners on a part, rather than drawing a box around the whole thing.
Text Link
iou
IOU
Intersection over Union: the overlap between a predicted box or mask and the true one, divided by their combined area. 1.0 is a perfect match.
Text Link
inverted-index
Inverted Index
An index that maps each term to the documents containing it, which is how full-text search answers queries quickly.
Text Link
instance-object-instance
Instance (Object instance)
One individual occurrence of an object in an image, as distinct from the class it belongs to.
Text Link
information-extraction
Information Extraction
Pulling structured facts such as names, dates, amounts or relationships out of unstructured text or images.
Text Link
indexing
Indexing
Building a lookup structure over data so searches find matches without scanning everything.
Text Link
image-processing
Image Processing
Operations applied to pixels, such as resizing, filtering, thresholding or colour correction, usually to prepare images for analysis.
Text Link
hyperparameter-selection
Hyperparameter selection
Choosing the settings that control how a model learns, such as learning rate, batch size or network depth, which are set before training rather than learned.
Text Link
grid-search
Grid search
Trying every combination of chosen hyperparameter values in turn and keeping the set that scores best.
Text Link
gradient
Gradient
The direction and size of change in a model's error for each weight, used during training to decide how to adjust that weight.
Text Link
gpu
GPU
Graphics Processing Unit: a chip built for many parallel calculations at once, which makes it the workhorse for training and running neural networks.
Text Link
generative-ai
Generative AI
AI that creates new content such as text, images, code or audio from a prompt, learned from patterns in large training sets.
Text Link
generalization
Generalization
A model's ability to perform well on data it has never seen, rather than only on its training examples.
Text Link
feedback-loop
Feedback Loop
A cycle in which a system's outputs, or human reactions to them, are fed back as training signal so it improves over time.
Text Link
feature-map
Feature Map
The output of one layer in a convolutional network: a grid showing where in the image a particular pattern was detected.
Text Link
feature-engineering
Feature Engineering
Creating or transforming input variables so a model can learn from them more easily, using domain knowledge rather than raw data alone.
Text Link
Keep reading

Tell me what you're building and what's in the way. I read everything.