Core Concepts
What Trakkr is, the vocabulary it uses, why your brand appears, the eight models, and the loop from measure to result.
On this page
Trakkr shows you how AI assistants answer the questions your customers ask, and whether your brand is in the answer. You choose the questions. Trakkr tracks recommendations across eight models and gives you a daily report of the available results.
That daily record is the product. Competitor rankings, citation lists, the work queue and the reports you send to clients all read from the same set of answers. This page is the vocabulary the rest of the docs assume. If you are setting up an account, start with Quick Start.
How AI answers differ from search
A search engine returns ten links and lets you choose. An AI assistant returns one answer and chooses for you.
When a model does not name you, you are invisible to a slice of buyers that web analytics will never show you. That gap is what Trakkr measures.
The vocabulary
Nine nouns explain most of what you see. The first six describe what Trakkr measures. The last three describe where you focus, what you do, and what keeps watch.
Brand
What you track. A brand is a name plus a domain, and it can carry aliases for the variations models use.
Prompt
A question worth tracking, written the way a customer would ask it.
Model
An AI engine. Trakkr polls eight: ChatGPT, Claude, Gemini, Perplexity, DeepSeek, Grok, Meta AI, and AI Overviews.
Mention
Your brand appearing in a collected answer. Position contributes to visibility; sentiment is measured separately.
Citation
A URL a model returned as a source. The clearest lever you have on what models say.
Competitor
Another brand tracked alongside yours. Powers every head-to-head view.
Page
One URL, moving through Available, Reached, Understood, Relevant, Selected, and Visited.
Action
A piece of work. Its work state stays separate from the result it later earns.
Automation
A standing instruction that checks something and responds inside the permission you set.
The daily report ties tracked prompts, model results, mentions and available citations together. Pages, Actions, and Automations turn that evidence into a path from finding a problem to measuring what changed.
Why you appear, or do not
When a buyer asks for the best CRM, these are useful areas to investigate. None is a rule that guarantees your brand will be named.
Training data presence
Review public coverage of your brand, including news, reviews and useful reference pages. You cannot infer which pages a provider trained on, or what a model knows, from a Trakkr score. Public source evidence is easier to inspect than assumptions about training.
Contextual relevance
Knowing you exist is not enough. The model has to connect your brand to the right category. "An all-in-one workspace" is harder to place than "a note-taking app for teams". Clarity wins.
Sentiment and positioning
How you are described matters as much as whether you are named. If the sources a model trusts position you as a leader, that follows you into the answer. If they position you as the alternative to someone else, that follows too.
Use Citations to inspect the sources models cite, then consider where useful public evidence about your brand is missing.
The eight models
ChatGPT
Claude
Gemini
Perplexity
AI Overviews
Grok
Meta AI
DeepSeek
Every tracked prompt goes to all eight, and Trakkr records the answer as it came back. Scoring 80 on Claude and 30 on Perplexity is normal. Both numbers are real, and they describe different audiences.
The models do not work the same way, and Trakkr does not flatten the difference. Each has its own training data, its own cutoff, and its own ability to look something up while answering. The part you can see in your own data is sources: some models hand back the pages behind an answer, and some answer from what they already hold and return nothing. Perplexity, AI Overviews and ChatGPT are the ones that contribute most of your citations, which is why a page you publish can show up in those first.
Treat each model as an audience rather than a tier. An improvement plan usually targets one or two at a time.
Traffic and crawler data
Two connections add data the models cannot give you. Google Analytics shows how many people reached your site from an AI assistant, which Visitors reports. A crawler tag shows which AI bots fetched your pages, and how many of those fetches happened inside a live conversation, which Crawlers reports. Neither is needed for the daily visibility report.
The visibility loop
Improving AI visibility is not a one-time audit. It is a loop, and the product is built around it.
- 1Measure
- 2Understand
- 3Act
- 4Result
| Stage | Where it happens |
|---|---|
| Measure | The Dashboard and Research set the baseline |
| Understand | Pages, Citations and Perception show what is holding you back |
| Act | Actions, Site Optimization, Content and AI Pages carry the work |
| Result | Results measures the finished work against the window fixed when it started |
The last stage has its own vocabulary, and it is deliberately careful. A measured result reads Earned, No movement, Reverted, Couldn't measure, or Coincided with a drop. None of those words claims the work caused the movement. Trakkr reports that a change and a movement lined up in time, in both directions, and leaves causation to you.
Where each idea lives
The rest of the docs are organized by what you want to do. Use this as a routing index.
| If you want to | Start here |
|---|---|
| See your numbers at a glance | Dashboard |
| Pick the right questions to track | Prompts, then Research |
| Find the page and stage that need work | Pages |
| Know where AI learns about you | Citations |
| Compare against rivals | Competitors |
| Know how AI describes your brand | Perception |
| Check whether AI states the facts correctly | Accuracy |
| See who and what arrives from AI | Traffic |
| Find content gaps | Content |
| Make your site easier for AI to read | Site Optimization and AI Pages |
| Ask questions of your own data | Agent |
| See what finished work earned | Results |
| Keep checking without doing it by hand | Automations |
For the formula behind any number, see Metrics. For a term you do not recognize, see the Glossary.
Common questions
Is AI visibility the same as SEO?
They overlap but they are not the same. SEO competes for a place in a list of links; AI visibility competes for a place inside a single written answer. Work that earns citations, clear positioning, and mentions in well-regarded sources helps both, but ranking first on Google does not guarantee a model will name you.
Do I need to install anything?
No. Adding a brand and a few prompts is enough for the daily model check. Two features do need a connection: AI traffic reads a Google Analytics property, and crawler tracking needs a small tag or edge function on your site.
How often do the numbers change?
Active brands receive scheduled daily reports. Check the report date and model coverage; a number can stay the same across reports. Longer jobs, such as a market scan in Research or a site audit, run when you start them.
Why is my score so different on each model?
Because providers, retrieval and response behavior differ, and available coverage can vary. A single blended number would hide the one fact you can act on, which is where you are weak. The model that matters most is the one your customers use, which is rarely the one with your best score.
Can I make a model say something specific?
No, and any tool that claims otherwise is guessing. You can change what the sources a model reads say about you, which is what Citations, Content, Site Optimization and AI Pages are for. The answer follows the sources, usually slowly.
What counts as a mention?
Your brand name appearing inside a model's answer to one of your tracked prompts. Trakkr uses ranked positions in successful collected answers to calculate visibility. Sentiment and brand descriptions are separate from that score.
Why does Trakkr never say a change caused an improvement?
Because it cannot know. Models change, competitors publish, and demand shifts, all inside the same window as your work. Results reports whether movement coincided with a change and holds the measurement window fixed so the comparison is honest, but it stops short of a causal claim.