Start here: three paths into AI visibility
A router for people new to AI visibility. Pick one of three paths — measure it, improve it, or choose a tool — see who this site is not for, and follow a realistic first week that ends with a baseline you can defend.
Path 1: measure it · Use the manual snapshot first
In short
This page routes you into RankRift based on what you are trying to do right now. If you need a number, start with measurement. If you already have a number and want it to move, start with the legitimate levers. If you are comparing platforms, start with the evaluation criteria before the product pages. Each path lists a first read, a first artifact, and a first action. The page also sets the site's dual register — plain English first, statistical and technical depth underneath — and names the readers this site is a poor fit for.
Who this is for
- An in-house SEO lead who has been asked whether the company shows up in ChatGPT and has no process for answering
- An agency strategist who needs to package AI visibility as a service without inventing a methodology overnight
- A founder or head of marketing who wants one concrete answer before deciding whether this deserves budget
- A technical SEO or engineer who has been handed a robots.txt decision about AI crawlers and wants a framework, not a default
- An analytics owner asked to prove that AI answers influenced pipeline
When to use this
- You are new to the category and want a sequence rather than a pile of articles
- You have a meeting this week and need vocabulary that survives questioning
- You have a vendor demo booked and want to know what to ask
- You want to know, honestly, whether this is worth your time yet
How to read this site
Every page opens in plain English and then goes deeper. The top of a page answers the question for someone with fifteen seconds and no background. Below that sit the formulas, the sampling logic, the provider documentation links, and the caveats.
That is deliberate. AI visibility is a statistics problem wearing a marketing costume, and most of the confusion in the category comes from people quoting a headline number without the sampling detail underneath it. Read the top for the answer; read the bottom before you present the answer to someone else.
Anywhere we report a measured figure, it ships with its prompt set, model versions, run dates, sample size, and variance, and links to the methodology.
Path 1 — Measure it
Choose this if you cannot yet answer the question factually. You do not need a tool, a budget, or a plan. You need a baseline and an honest confidence range around it.
The sequence is: understand why answers vary, build a prompt set that reflects how buyers actually ask, decide how many runs you need, then record the result in a form your future self can compare against.
- Read the measurement hub — metric definitions, sampling, prompt-set design
- Follow the manual snapshot method for a first sampled reading
- Use the sample-size calculator to plan how many runs a real test needs
- Build a prompt set that reflects how your buyers actually ask
- Measure AI visibility in a spreadsheet before you buy anything
Path 2 — Improve it
Choose this if you already have a baseline and the answer was uncomfortable. The work splits into two halves that people routinely confuse: making your content retrievable at all, and making your brand the kind of thing an assistant has reason to name.
Expect uneven, slow, and partially unattributable results. There are two clocks running — the retrieval clock, which can move in weeks, and the training clock, which does not respond to anything you did last Tuesday. Anyone promising a fast, guaranteed change in an AI recommendation is selling something we would advise you not to buy.
- The levers that legitimately change AI recommendations
- Technical retrievability — crawlers, rendering, schema, structure
- Entity clarity: making sure AI knows what you actually are
- How long it takes to change an AI recommendation
- The bright lines: what not to do, and why each backfires
Path 3 — Choose a tool
Choose this if you are comparing platforms or defending a purchase internally. Start with the criteria rather than the shortlist: which models a tool covers, how it collects data, whether its metric definitions are published, how often it runs, and what it exports.
We write about trackers without operating or selling one. The disclosure page states the rules for vendor coverage, and the evaluation criteria come before any product mention. Every comparison should name who a different approach serves better; sometimes the honest answer is that you do not need a tool yet.
- Tool evaluation criteria, with the conflict stated up front
- Build, buy, or spreadsheet — including not-yet as a valid answer
- Twelve questions to ask any vendor about their methodology
- How tool coverage stays independent
A realistic first week
This assumes a few hours total, spread across five days, with no purchase. It ends with a documented baseline rather than a strategy deck.
- Day 1 — Use the manual snapshot and read the result including its stated limits
- Day 2 — Learn the vocabulary so the numbers mean something specific to you
- Day 3 — Draft a prompt set of buyer-realistic questions, avoiding any that name your brand
- Day 4 — Decide your run count and record models, versions, dates, and raw responses
- Day 5 — Write down the baseline and one hypothesis about why it looks the way it does
Do not act on week-one results. A first measurement tells you where you stand, not what changed or why.
Who this site is not for
Being explicit about this saves everyone time.
- Anyone looking for guaranteed placement in an AI answer. No such thing exists, and we will not pretend otherwise.
- Anyone wanting instructions for astroturfing, incentivised reviews, parasite pages, or prompt injection. We cover these defensively — what they are, how they are detected, what they cost — and never as how-to.
- Teams with no measurement capacity at all. If nobody can run a prompt set monthly, a tracker will not fix that.
- People who want a single vanity score with no confidence range. Any number we publish comes with error bars, and sometimes with the word inconclusive.
- Anyone wanting measurement by scraping assistant interfaces or automating consumer accounts. We use official APIs only and do not describe alternatives.
What to skip in your first month
Attention is the scarce resource, and the category is loud. A few things reliably consume a month and return nothing measurable.
Skip per-model micro-optimisation before you have a cross-model baseline. Skip any single tactic presented as the one thing that works. Skip building a dashboard before you know which metric answers your actual question. And skip arguing about acronyms — the taxonomy post covers which of them describe real practice.
- AEO, GEO, LLMO, AI SEO: which acronyms actually mean anything
- Chasing quick fixes while ignoring the basics
Related pages
Frequently asked questions
Which path should I pick if I am not sure?
Take the measurement path. Almost every question in this category resolves into whether you can produce a number you would defend under questioning. Improvement work without a baseline cannot be evaluated, and tool selection without knowing what you need to measure turns into a feature beauty contest. If the baseline turns out to be fine, you have saved yourself a quarter of unnecessary work.
How much time does a first baseline actually take?
For a small prompt set across a couple of models, expect a few hours of setup and a modest amount of API cost, plus a shorter re-run each month. The expensive part is not the running, it is deciding what to ask and writing down what you did so the next run is comparable. The measurement hub walks through that decision, and the calculator helps you size the run count.
Do I need a budget before any of this is useful?
You need enough API credit to run a prompt set a few dozen times, which is small. You do not need a platform subscription to start. The spreadsheet method is a genuine method, not a consolation prize, and it stays viable until the number of brands, categories, or models you track makes manual runs unmanageable. That crossover point is exactly what the build-buy-or-spreadsheet guide is about.
Is it too early to care about this in my category?
Possibly, and we would rather you find that out cheaply than be scared into a purchase. Use the manual snapshot. If assistants barely engage with your category, or if the answers are unstable to the point of noise, note that, set a reminder to re-run in a quarter, and spend your effort elsewhere. Recording a dated null result is a legitimate outcome and gives you a comparison point later.
What should I read before a vendor demo?
Read the tools hub for the evaluation criteria and the twelve methodology questions, then ask any vendor how they collect data, which model versions they query, how many runs sit behind a reported figure, and how they express uncertainty. Also ask what happens to your prompt sets and results. Answers that are vague on collection method or sample size are the ones worth pressing on.
Why does this site keep saying it does not know something?
Because the honest state of the evidence in this category is thin, and confident claims about model internals are mostly guesses dressed up as expertise. Where provider documentation exists, we link and date it. A measured claim appears only with its published run. Everywhere else we write the uncertainty explicitly, which is less satisfying to read and considerably safer to act on.
Pick a path and get a baseline this week
The manual snapshot guide gives you a sampled starting point, and the glossary gives you the language to explain it. Everything else can wait until you have a number.