AI Visibility Without an SEO Background: Guided Mode, Tours, and the Assistant

Short answer: CiteCue ships two dashboards from the same data. Guided mode reduces the product to four places and one next action; Expert mode gives you every page at once. Interactive tours explain the interface in place, and an assistant grounded in your own scan data answers the questions a tooltip can't.

The uncomfortable truth about our category is that most of the tools in it — ours included, in its first version — are built for someone who already speaks the language. They open on a grid of metrics named share of voice, citation share, mention position, and quietly assume you know which one to care about on a Tuesday morning.

Sometimes that person exists. Often the work lands on a founder, a marketing generalist, or the one person at an agency who drew the short straw on "the AI thing." They don't need a course in answer engine optimization. They need to know what to do next.

So we built the product twice.

Two dashboards, one dataset

There's a Guided / Expert switch in the dashboard. It's a per-user setting, not a plan tier and not a permanent decision — you can flip it as often as you like, and nothing is hidden from you that a click doesn't bring back.

Expert mode is the full sidebar: prompts, citations, competitors, platforms, sentiment, brand risk, insights, readiness, agent traffic, referrals, trends, the report, and every settings page. If you know what you're looking for, it's the fastest way there.

Guided mode is four items:

Settings — brand context, integrations, members — collapse into one group at the bottom. Same scans, same scoring, same data underneath. Guided mode isn't a lite version; it's a smaller doorway into the same house.

The Home page answers three questions, in order

Guided Home opens with a progress rail, and the rail changes depending on where the project actually is.

Before your first scan it reads Confirm brand → Review questions → Run first scan, and it derives each step's state from real project state — whether we've finished reading your site, how many questions are switched on, whether a scan is running right now. There's no checklist you can tick without doing the thing.

After results land, the rail becomes Detect → Fix → Prove, which is the whole loop the product exists to run:

• Detect — the questions buyers might put to an AI assistant about brands like yours. CiteCue generates a starter set from your site's brand context; you keep the ones that match how people really discover you and add your own.
• Fix — the concrete content changes that should move the number.
• Prove — the next scan, which confirms whether they did.

Underneath the rail, the headline is a sentence rather than a metric: AI mentions your brand in 14 of 60 answers. The 0–100 visibility score is still there, with a letter grade and its change since the last scan, and next to it a plain comparison — your cited pages versus your closest competitor's. But the first thing you read is a fact in English.

One next best action, not a to-do list of forty

Then comes the part we argued about most: Your next best action. One card. The highest-impact open fix, its impact label, and a "why it matters" paragraph in non-specialist language. Below it, a short numbered list of other things worth doing, and a link to the full queue if you want it.

Where a fix is one CiteCue can apply for you, the AI Auto-Fix button sits right on the card. Where it isn't, the fix opens into a platform-aware, step-by-step guide — the instructions differ depending on whether you're on WordPress, Shopify, Webflow, or something bespoke, because "add JSON-LD to your homepage" means five different sequences of clicks.

One detail we added after watching people use it: when you apply a fix, it leaves the queue. That looked like lost work. So the card now says "3 fixes are applied and waiting for your next scan to confirm them" — and it says it loudest when the queue is empty, because an empty queue you earned should not look identical to an empty queue you never started.

Tours that point at the actual screen

Documentation in a separate tab is documentation nobody reads. CiteCue's tours run on top of the live interface and spotlight the real element.

There are three kinds:

• A first-run journey that walks across pages — your brand context, then your questions, then your first scan — navigating for you between steps.
• A guided-dashboard tour that explains Detect, Fix, and Prove where they sit on Home.
• Per-page mini-tours for the denser screens: prompts, prompt detail, citations, competitors, content fixes, readiness, and AI context.

Two behaviors matter more than the content. Steps are conditional: the "add your first prompt" step only appears if you actually have zero prompts, and it waits for you to add one before continuing. And steps whose target isn't on screen are skipped, not faked — a tour never spotlights an element that isn't there or narrates a feature your plan doesn't include.

You can quit any tour instantly, and restart it from the Guide page, which doubles as a plain-language index of every page in your sidebar, what each metric means, and how a scan works end to end. Metric names throughout the dashboard also carry a small ⓘ that explains them in one sentence, in the place where you're confused rather than three clicks away.

An assistant that can only tell you true things

The tours cover where things are. The assistant covers what your numbers mean.

It opens as a side panel that deliberately doesn't block anything — no dimming, no trapped focus. You can keep clicking around the dashboard with it open, and it follows you: the suggested questions change per page, so on Citations it offers "Where do competitors get cited but I don't?" and on Competitors it offers "Who is winning share of voice?"

The design constraint we set for ourselves is that it must not be able to make things up:

• Every number comes from a tool call against your project's data — overview, competitors, prompts, citations, ranking-factor gaps, sentiment, referrals, opportunities, report summary. It cannot answer a data question from memory, and if a tool returns nothing it says so.
• You can audit the work. Each answer carries an expandable "Ran 3 tools" trace listing exactly what it queried and what came back.
• With no completed scan, it refuses to speculate and tells you to run one instead.
• It stays in scope. Questions about your AI visibility, about how CiteCue works, and about improving your visibility get answered. Requests to write unrelated code or do your homework get one polite sentence back.

There's also a small ✦ next to metrics, fixes, and prompts across the dashboard — Ask about this — which opens the assistant already pointed at that specific item, so you never have to describe what you're looking at. Conversations are saved, so the question you asked last week is still there.

What this doesn't do

Three honest limits.

The assistant is a paid feature. Free projects don't get it; Pro includes 100 messages per project per month and Agency 1,000. That's a cost decision about real model calls, not a growth tactic, and the pricing page states it plainly.

Guided mode genuinely hides pages. Sentiment, brand risk, agent traffic, trends, and the topic grid aren't in the four-item nav. If you want them, Expert mode is one click, and nothing you did in Guided mode is lost or reset.

And tours are best-effort on small screens — a step anchored to the desktop sidebar will skip on a phone rather than point at nothing.

Why we spent the time here

The category's real failure mode isn't inaccurate measurement. It's a dashboard that's correct and unread — a score that drops, a notification nobody acts on, and a renewal conversation where the honest answer is "we never really used it."

Guided mode, the tours, and the assistant are all answers to the same question: what should this person do next, and do they understand why? Content Fixes decides what to work on and Prompts Monitoring tracks whether it worked — but neither matters if the person who owns the work can't find their way in.

If you want to see it, the fastest path is to create a project and let the first scan run. The rail will tell you where you are, and the assistant is right there when the numbers stop making sense.