Visibility

Overview#

Visibility answers a single question: when somebody asks an AI assistant something in your category, does it mention you at all?

The whole screen is built around weeks rather than individual runs, and that choice shapes how everything on it should be read. Any single run varies on its own for reasons that have nothing to do with you, so comparing one run against the next invites conclusions the data cannot support. A week gathers up several runs, which is enough for a change to mean something.

Discovery Rank, and why lower is better#

The headline number here is Discovery Rank, and it catches almost everybody out at least once because it runs in the opposite direction to most numbers in the product. Seerly explains it in a tooltip on the column header:

Your brand's position among all competitors, normalized to a 1-10 scale. Ranking is based on mention frequency (presence), average position in AI responses, and share of voice to break ties. Lower is better. A rank of 1 means you're the top-ranked entity. A rank of 10 means you're last among all competitors.

So a rank moving from 6 to 3 is good news, not bad. Because it is a position relative to your competitors rather than a percentage of anything, a rank can also get worse during a week when nothing about your own presence changed at all. Somebody else simply improved, and you moved down the order without moving yourself. That relativity is worth holding on to, since it explains most of the confusing weeks you will have with this screen.

Panels and controls#

Discovery Reach Trend charts the direction over time, and it is the right thing to look at first. A single week tells you where you stand; the line tells you whether any of the work you have been doing is landing. Weeks with no runs are left out, so the line joins the weeks that were measured instead of dropping to zero, and the count above the chart says how many weeks had runs.

Share of Voice sits alongside the rank and feeds into it, so reading the two together tells you whether a rank change came from your own movement or somebody else's. There is more on this in Competitors.

The ring and the number in its middle answer different questions, which catches people out. The ring divides a full circle between you and your tracked competitors, so it describes how you compare to them. The number in the middle is your share of all citations, including every source that is not in your tracked set at all. A brand can therefore fill a perfectly healthy looking wedge of the ring while the centre reads a low single-digit percentage, and both numbers are correct.

Read the ring for your standing against named rivals and the centre number for how much of the whole conversation you actually hold. A wide wedge sitting alongside a small centre number usually means your category's answers are dominated by sources you are not tracking, which is worth following into Citations.

Share of Voice

Brand citations vs competitors

4.3%

Your brand

• your brand(4.3%)
• competitor one(2.3%)
• competitor two(1.8%)
• competitor three(1.0%)

The Trend filter is a button labelled Trend on the filter bar, and it is the most commonly missed control on this screen. Its options live inside the dropdown rather than on the table itself, so hunting along the column headers for something to sort by will never turn them up. Clear trend appears once a filter is applied, and Reset clears everything back.

Trend

Trend

Trend

• All trends
• Gainers
• ✓Decliners

There is a Search topics... box for finding a specific topic, and Columns or Toggle columns for showing and hiding what is displayed.

Recommendation Rate sits beside Mention Rate once recommendation tracking covers the week on display for your project. Mention Rate counts every answer in the latest week that names you. Recommendation Rate counts only the answers that put you forward as an option. A wide gap between the two means AI engines know your brand but are not suggesting it. The change against the week before is shown underneath, in percentage points. The column is calculated for your own brand only, so it is not shown while a competitor is selected, and a dash on a single topic means that topic's figure is not available yet.

Weekly drivers is what to reach for when you have twenty topics and twenty minutes. It splits into Topics gaining ground in the latest week and, under the Needs attention heading, Topics losing ground in the latest week. Starting here rather than scanning the full list is almost always the better use of the time, because the panel is the screen answering "what changed?" so that you do not have to work it out yourself.

A topic you were not ranked for in the previous week is labelled Surfaced rather than being given a number, and one you have stopped ranking for is labelled Dropped. Neither has two positions to measure a movement between, so the card names the change instead of counting it, and a week holding no position at all reads absent in place of a rank. Topics that ran no prompts at all in the previous week are left out of the panel entirely, because there is nothing to compare the latest week against.

Each card also carries a mention rate, the share of that week's runs which mentioned you at all, shown against the previous week's figure. Read it alongside the rank rather than as a second version of it, because the two can genuinely disagree. Rank only describes the runs you appeared in, so a week can rank you better on fewer appearances, and the pairing is what tells you which of those happened.

Weekly drivers

Topics gaining ground in the latest week

project management for agencies, up 3 positions

Needs attention

the heading over the week's losses

Topics losing ground in the latest week

client onboarding, down 2 positions

Surfaced

consulting retainers, absent the previous week

mention rate

20% to 43%

Prompts Ran and Prompts included in the latest week tell you how much data the week is actually resting on, and they are worth a glance before you draw any conclusion from a movement. A topic with only a handful of prompts behind it in a given week is a much thinner reading than one with many, and that difference is completely invisible in the rank itself. Competitors and Top competitors for this topic show who else is holding the answers.

The change indicators are deliberately explicit rather than being arrows you have to interpret. Now mentioned (was absent before) and No longer mentioned (was present before) mean what they say. Not mentioned in any prompt/run is the one to read carefully, because it is a different thing from a poor rank: the topic produced no mention of you anywhere, rather than a weak position. You may also see No topics found, or Absent where you did not appear.

Reading a topic in detail#

Opening a topic shows you the prompts inside it, and this is where the screen stops summarising and starts being useful.

A topic's rank is an average, and averages hide the thing you can actually act on. A middling rank is rarely a case of being moderately visible everywhere. Far more often it is real strength on two prompts and complete absence from three others, and those three are the work.

The prompt table gives you the Prompt itself as it was asked, Lifetime Mentions for how often you have been named across all runs, Recommended (shown together with Recommendation Rate) for how often the answer recommended you, Lifetime Runs for how often the question has been put, and Last Run for when it was last asked. Unlike the lifetime figures beside it, Recommended starts from the date recommendation tracking began for your project, and hovering the column heading shows that date once tracking has begun. Because it covers a shorter period, it can sit above Lifetime Mentions for a prompt that has been mentioned more often lately.

Prompts inside a topic

Prompt	Lifetime Mentions	Recommended	Lifetime Runs	Last Run
What's the best project management tool for a small agency?	12	25%	40	2 days ago
Which project management tool is easiest to onboard a team onto?	0	0%	38	2 days ago
Is paid project management software worth it for a team of five?	7	10%	40	2 days ago

Those lifetime figures carry more weight than they first appear to. A prompt asked many times that mentions you rarely is a settled weakness and a reasonable thing to spend effort on. A prompt asked only a handful of times is not yet evidence of anything either way, and treating it as though it were will send you chasing noise.

Select a prompt to list its runs, then select a run to read that engine's full answer under Response. A Recommended card above it reads Yes when the answer puts your brand forward as an option and No when it does not, including when your brand is not mentioned at all. It reads Not available when the answer was not checked or could not be judged. For a run from before recommendation tracking started for your project, the card says from which date it is available. Directly under the answer, ChatGPT and Perplexity runs list the Query fan-outs, the searches the engine reported running to build it. Google does not publish the searches behind an AI Overview, so Google AI Overview runs show Related searches and People also ask from the same Google results page instead. Each heading carries a count of its entries. A run shows these lists when its engine reported them. Runs from before the lists were added show none.

The list of runs shows ten at a time, with page numbers beneath it. A date range narrows the list, and a prompt that has run on more than one engine also offers an engine filter.

For the week as a whole, a prompt that has any shows a Query fan-outs button beside its text, with the number of entries captured that week. Select it to open a panel that groups that week's entries by engine: the searches ChatGPT and Perplexity reported, and Related searches and People also ask for Google AI Overview. An entry that came up more than once shows how many times. A prompt with none that week has no button. The panel opens on the week the dashboard is showing. Arrows beside the week move to the nearest earlier or later week that has entries for that prompt, and an arrow appears only when there is such a week.

Top competitors for this topic shows who is turning up where you are not, which is the quickest route from a weak topic to something specific you can go and read. Each company appears by name with its domain under it, as it does in the Share of Voice legend. If a topic has no prompts against it you will see No prompts found for this topic.

Best practices#

Because Discovery Rank is relative, there is no absolute number to aim for. A rank of 4 in a crowded category can represent a considerably stronger position than a 2 in a category with three players in it. Judge it three ways instead:

• Moving down across several weeks, since lower is better here. Steady improvement tells you far more than any single reading.
• Spread evenly across topics. A respectable average concealing one commercially important topic sitting at rank 10 is a worse position than a moderate average distributed evenly.
• Backed by enough prompts for the week to carry any weight at all.

Next steps#

1. Read Weekly drivers first, since it tells you what moved without you scanning anything.
2. Open the topics flagged Needs attention, along with any that dropped.
3. Set the Trend filter to Decliners when you want the full list of what moved against you rather than only the highlights.
4. Check Prompts included in the latest week before concluding that a drop is real, because a thin week is a thin reading.
5. Open the topic itself and find the specific prompts you are absent from. This is the step that turns a rank into something you can act on.
6. Look at Top competitors for this topic to see who is holding the answer in your place.
7. Follow it into Citations to find what is actually being cited. Visibility tells you that you are missing; citations tell you what is winning instead.

What to keep in mind#

Discovery Rank measures your position among competitors inside AI answers. It says nothing about traffic, conversions or revenue, and it was never meant to. Analytics is where that connection gets made.

It also covers only the engines your plan tracks and the prompts you have configured. A question that no prompt represents is simply not measured, and an unmeasured gap looks exactly like no gap at all from here, which is why prompt coverage deserves as much attention as any rank does. See Prompts.

Finally, it does not tell you why you are absent. That is the job of Gap analysis and Citations.