Mastering Competitor Benchmarking

There are two ways to measure a competitor here, they answer different questions, and confusing them is the main reason people misread this page.

The first is co-mention: when an engine answers your prompts, which other brands does it name alongside you? That comes free with every scan you already run. The second is a first-class competitor scan: run the same prompt set as if you were measuring them, and get a real visibility score to put beside your own. That costs credits, because it is a whole scan.

This tutorial sets up both and then shows you what each one can and cannot support.

Step 1: Decide who is actually worth tracking

Competitors here are brand names, matched against what engines say. Choose them accordingly:

• Direct rivals — the companies you lose deals to. Always track these.
• Category incumbents — the names an engine reaches for by default when asked “best X”. They may not be your commercial competitors at all, but they occupy the answer you want.
• Adjacent substitutes — the “you could just use a spreadsheet” answer. Frequently named, rarely on anyone’s competitor list, and the reason a category question does not produce a shortlist you are on.

Keep the list short. Every name you add appears in the co-mention table, and every name you run a first-class scan against costs another full scan’s worth of credits. Five well-chosen names beat twenty — which is also the hard ceiling: a project accepts 20 competitors, the same figure on every plan, and the twenty-first comes back as a constraint error rather than a silent truncation.

Step 2: Add them to the project

Open Competitors in the Monitor group of the sidebar. The first card, Tracked Competitors, takes a competitor name and saves instantly — there is no separate save step and no domain field. The list belongs to the active project, so a workspace tracking several brands keeps a separate roster for each.

Use the name a person would type into a chat box. Engines answer in prose, and the matching is against that prose, so the trading name usually works better than the legal entity or a URL.

Step 3: Read the co-mention stats from your existing scans

Four stat cards sit below the setup cards: Your Mention Rate, Competitors Seen, Rivals You Lead and Top Rival. Underneath, the Share of Voice table lists each brand with how often it appeared across your tracked queries.

These all come from the answers to your prompts in your latest completed scan, which has two consequences worth stating plainly:

• The denominator is your prompt set. A rival that dominates a market you do not ask about will not appear. This measures the conversation you chose to measure.
• “Competitors Seen” is bounded by your tracked list. The analyser is asked which of the competitors you configured were named in an answer, and anything else it returns is discarded before storage — so this counts how many of your rivals showed up, never who else did. Discovering a rival you have not added is a manual read of the answers, not something this card surfaces.

“Rivals You Lead” compares your pooled mention rate against each competitor’s over the same answers. It is a coarse measure and it is honest about being one: it counts appearances, not recommendations. A brand mentioned in every answer as the thing to avoid leads on this metric.

Step 4: Run a first-class scan for score-vs-score

Co-mention counts cannot tell you how well a competitor is doing — only how often they came up in your answers. The Competitor Benchmark table fixes that. Each row carries a Scan button that runs the same prompt set with that competitor as the subject, and returns a real score, mention rate, share of voice and average rank for them, sorted highest score first.

Three things to know before you click it:

1. It costs what a scan costs — one credit for every prompt that comes back with an answer, checked across every engine you have enabled, every time you press it. The engine count does not multiply that, but the competitor count does: benchmarking four rivals is four scans. Do it deliberately, on a cadence, not reflexively.
2. It notifies nobody. A competitor scan produces no email and no webhook, on purpose: a digest reading “Competitor AI visibility 74/100” delivered to your team as if it were your own result would be actively misleading. Read competitor results in the dashboard.
3. It is only comparable because the prompts are identical. That is the whole design. If you change your tracked prompts between two benchmark runs, the comparison is void — the instrument changed, not the market.

Step 5: Read the share-of-voice trend, gaps included

Once you have at least two completed scans, a Share of Voice over time chart appears: your share of brand mentions versus everyone else’s, per scan.

The detail that matters is what happens when a scan mentioned nobody — not you, not a competitor. The chart draws a gap at that point rather than a zero.

That is deliberate and it prevents a specific, damaging misreading. A scan with an empty denominator has no share to report. Plotting it as 0% would assert that competitors held 100% of the conversation, which is exactly wrong and exactly what a new project’s early scans look like when nobody in the category was named yet. A gap says “no data”; a zero would have said “you lost”.

So when the line breaks, look at the scan behind it before you react.

Step 6: Turn the comparison into work, and know what not to conclude

The useful output of this page is a list of prompts where a rival is recommended and you are not. Cross-reference the Content Gaps view for exactly that set, then decide per prompt:

• They are named, you are absent. A coverage gap. Something exists — their own page, a roundup, a review — that the engine can reach and nothing equivalent exists for you.
• You are both named, they are first. A positioning gap. Publish something that answers that specific question directly rather than describing your product in general. Google’s guidance on creating helpful, reliable, people-first content describes the same standard from the search side, and it holds here: the page has to answer the question, not gesture at the topic.
• You lead and they are absent. Do nothing. This is the row to leave alone.

Three conclusions this data does not support, and all three get drawn anyway:

• “Their share of voice fell, so they lost visibility.” Share of voice is relative. It also falls when a third brand rises.
• “We are ahead of them.” On your prompts, in the markets you configured, on the engines you selected. Say so out loud when you present it.
• “The gap moved, so our work is paying off.” One scan is one sample. Look at the trend across several before claiming causation — and re-scan on the same prompt set, or you have measured the change you made to the measurement.

Step 7: Where to go next

• Competitor benchmarking — the feature page behind this view.
• Share of voice explained — what the metric does and does not mean.
• How to do an AEO competitor analysis — the wider method this page slots into.
• Why is my competitor in AI and I’m not? — the diagnosis for the most common finding here.
• Setting Up Tracked Prompts — the shared prompt set is what makes every comparison on this page valid.