The AI-ops stack,
in plain words.

How assistants choose brands, and the metrics for whether they choose yours.

AI visibility

Whether — and how accurately — AI assistants like ChatGPT, Perplexity and Google AI name and describe your brand when a buyer asks for a recommendation in your category. The AI-era successor to a search ranking: if the assistant answers the question for the user, being on page one of a results list no longer matters.See → AI Brand Visibility

Share of modela.k.a. AI share-of-voice

The share of relevant AI recommendations in which your brand appears, versus competitors. The AI-era analog of share-of-search or share-of-shelf: if an assistant names three brands when asked for the best option in your category and you are one of them across a third of the prompts a real buyer would use, your share of model is roughly 33%.

Recommendation gap

The set of high-intent buyer questions in your category where an assistant recommends competitors and never mentions you. It is the absence that costs the sale — the customer gets a confident shortlist of two-to-four brands and you are simply not on it. Measuring it is the first job of an audit.See → AI-Visibility Teardown

Generative Engine OptimizationGEO

The practice of structuring a brand's information so generative AI engines cite it correctly — the AI-era counterpart to SEO. Where SEO optimized for a ranked list of blue links, GEO optimizes for being the source an assistant draws from when it writes the answer. In practice it means machine-legible facts, structured data and answer-first content, not keyword density.

Answer Engine OptimizationAEO

Optimizing content so an answer engine can extract a clean, quotable answer from it directly — typically via answer-first writing, FAQ and DefinedTerm structured data, and tables. Closely related to GEO; AEO emphasizes the page-level format that makes a passage liftable into an answer, while GEO is the broader brand-legibility discipline.

The machinery that decides what an AI model knows about you — and whether it is right.

RAGRetrieval-Augmented Generation

A pattern where a language model retrieves relevant external text at question time and writes its answer from those passages, instead of relying only on what it memorized in training. RAG is how you ground a model in your own up-to-date information — but plain RAG retrieves text that looks similar, which is not the same as text that is true about you.

GraphRAG

Retrieval-augmented generation that walks a knowledge graph — entities and the explicit relationships between them — instead of fetching similar-looking text chunks. It answers multi-step questions (this product, at this dose, with this contraindication) from connected facts, so the answer is correct rather than merely plausible. The defensible layer under AI visibility for catalogs and regulated categories.See → Brand Ontology & GraphRAG

RAG vs GraphRAG

Traditional RAG retrieves similar text by vector match and lets the model stitch an answer from passages; it has no model of how facts relate. GraphRAG retrieves connected facts by traversing a graph of typed entities and edges, so it handles multi-hop, relational questions accurately. For brand-specific answers, that is the gap between fluent and correct.

Brand ontology

A structured model of your brand as entities and relationships — products, ingredients, dosages, claims, certifications, differentiators and the links between them — that a machine can read and reason over, rather than the marketing prose and PDFs an AI engine otherwise has to guess from. It is the substrate GraphRAG retrieves from, and an asset you own outright.See → Brand Ontology & GraphRAG

Knowledge graph

Data stored as a network of typed entities (nodes) connected by explicit relationships (edges), rather than as rows in a table or free-form text. Because the relationships are first-class, a system can traverse them to answer questions that span several connected facts — which is exactly what makes a brand's catalog legible and answerable to AI.

Embedding

A numeric vector that represents the meaning of a piece of text, so that passages with similar meaning sit close together in vector space. Embeddings power vector search and plain RAG: the engine finds passages near your question. Useful for similarity, but blind to whether two near-neighbors are actually factually related — the limitation a knowledge graph removes.

Hallucination

When a language model states something fluent, confident and false — because it optimizes for plausible-sounding text, not for truth, and has no built-in fact-check step. For a brand, the dangerous form is a confidently wrong answer about your product, price or claims. Grounding the model in a structured, sourced knowledge layer is the architectural fix.

The software that does the work — and what it means to run it, not just ship it.

Agentic workflow

An AI system that pursues a goal across multiple steps — planning, calling tools, reading results and deciding what to do next — instead of producing a single one-shot response. Agentic workflows are how AI does real operational work (research, outreach, reconciliation, content) rather than just answering a prompt. They need guardrails and a human-in-the-loop where stakes are high.See → Agent Workflows

Multi-agent system

An architecture where several specialized agents collaborate on a task — for example a planner, a worker and a reviewer — each scoped to one job, coordinating through shared state or messages. The pattern trades the brittleness of one giant prompt for the reliability of small, checkable, separable roles.

Digital employee

An operated AI agent that owns a recurring business function end to end — a content engine, an inbound qualifier, an ops reconciler — running on a schedule with monitoring and a human approving the consequential moves. The framing matters: you are not buying a tool you must operate, you are getting a function that is run for you.See → Operating Partner

AI operating layer

The standing layer of AI systems that runs across a company's functions — visibility, content, support, operations — built on a shared brand ontology and operated continuously, rather than a pile of disconnected point tools. PlanePaper's core idea: AI you operate, not AI you ship and abandon.

Human-in-the-loopHITL

A design where a person reviews or approves an AI system's consequential actions before they take effect — sending an email, publishing a page, moving money. It is how you get the leverage of automation without surrendering judgment on the decisions that carry real risk. The default posture for anything client-facing or irreversible.See → How we work with data

Eval harness

A repeatable test suite that scores an AI system's outputs against known-good answers and failure cases, so you can measure quality and catch regressions before they reach production — the AI equivalent of unit tests. Without one, you are shipping on vibes; with one, you can change a prompt or model and prove it did not get worse.

Prompt engineering

Designing the instructions, context and examples given to a language model to get reliable, on-task output. It is a real and load-bearing craft for early prototypes, but on its own it is the weakest layer of an AI system — durable reliability comes from grounding, evals and architecture, not from a cleverer wording of the request.

What the words on our products pages actually mean.

AI Brand Audit$1,500 · the starting point

PlanePaper's flat-fee diagnostic: a week-long teardown that measures how AI search describes your brand versus competitors in hard numbers, diagnoses the architectural cause of any gap, and hands you a roadmap of systems ranked by ROI that is yours to keep — regardless of whether you build with us. The honest front door to every engagement.See → AI Brand Audit

AI-Visibility Teardownfree · the warm-up

A free, fast snapshot of how ChatGPT, Perplexity and Google AI describe your brand today versus competitors — whether you are named, how you are described, and who wins the answer instead. The no-commitment warm-up to the full $1,500 audit; you keep the snapshot either way.See → Get the Teardown

System Sprintfixed scope · build

A fixed-scope engagement that takes one prioritized system from the audit roadmap and ships it to production — a brand ontology, a visibility engine, a content engine or an agent workflow — built properly rather than prototyped. The bridge between knowing the fix and having it live.See → System Sprint

Operating Partnermonthly · run

An ongoing engagement where PlanePaper runs and grows the AI systems it built — monitoring them, extending the ontology, and operating the digital employees — so the work compounds instead of decaying after launch. The 'operate' half of build-and-operate.See → Operating Partner