How do I benchmark my brand's AI citations against competitors?

Lock and date a prompt set grouped by buyer-journey stage, then run it on each engine your buyers use, several times per prompt, in a clean session. For every run, record which brands are mentioned and which URLs are cited. Compute citation rate and share of voice per brand and per engine, then open the competitor URLs that win citations to see what those pages cover.

What is a good AI visibility score?

No universal benchmark exists. Each tool builds its score from its own engines, prompts and aggregation, so a number from a given tool cannot be read against another tool's. The useful benchmark is relative: your mention rate, citation rate and share of voice against named competitors, on the same locked prompt set and engine. A good result is a lead over the rivals that matter on the prompts closest to a purchase.

How many times do I need to run a prompt to get a reliable comparison?

A single run is noise, because answers vary between runs. Our working rule is several runs per prompt per engine in each period, spread across different days, with mention rate counted as the fraction of runs that name a brand. Treat a gap no larger than a single run as a tie, and confirm a trend in the following period before acting on it.

Which AI visibility tool can track multiple competitor brands at once?

Judge tools on criteria rather than names. Check how many brands a plan allows per project, because your competitor set is the list it must track. Confirm it covers the engines your buyers use, reports each engine separately, and gives access to raw mention and citation counts and archived answers. Without raw data, you cannot normalize its share of voice or audit a gap.

Why does my AI visibility score change even when my citations stay flat?

Many scores are shares or composites, so they move when anything in the calculation moves. Prompts or brands added to the set change the denominator. A new engine or model version changes the answers. A rival can also gain mentions without citations, which lowers your share of voice while your citation count stays the same. Check these causes before reading the change as a gain or a loss.

How to Compare Your AI Visibility Against Competitors

To benchmark your AI visibility against competitors, run the same locked, dated prompt set on the same engines for every brand you track. Repeat each prompt enough times to separate a real gap from answer variance. Then compare raw metrics per engine (mention rate, citation rate, share of voice) rather than proprietary scores that tools compute differently.

What benchmarking AI visibility against competitors actually measures

A competitor benchmark counts the same measures for every brand in the set, on identical prompts, per engine. A standalone score for your brand tells you little; the comparison is the result.

Start with a distinction worth drawing precisely:

A mention is your brand name in the generated answer.
A recommendation is a mention presented as a suggested choice, not a passing reference.
A citation is a URL of yours listed as a source for the answer.

From these, the benchmark derives per-brand measures: mention rate, citation rate, and share of voice, your mentions as a share of all mentions of the tracked brands. Also record position in the answer, recommendation versus passing mention, and whether the description is accurate.

Track each engine separately: ChatGPT, Perplexity, Gemini, Google AI Overviews and AI Mode, plus Claude where your buyers use it.

Google rankings are no proxy for these answers. In a study published on 11 August 2025, Ahrefs reports: "In a dataset of 15,000 prompts, we found that—on average—only 12% of links cited by ChatGPT, Gemini, and Copilot appear in Google's top 10 results for the same prompt." The answers themselves have to be measured.

Citations make the benchmark actionable: a competitor's cited URL is a page you can open, study and compare with your own coverage of the question.

How to choose the competitor set and the prompt set for the benchmark

The benchmark rests on its inputs: the brands and the prompts. Choose both deliberately and write them down before you measure anything.

Who the engines put next to you: co-mentioned, compared and cited

Build the competitor list from the brands you would name yourself, then add the brands the engines surface on category prompts. Sort them into groups:

Co-mentioned brands appear in the same answers as you.
Compared brands are the ones an engine sets against you in comparison or alternatives answers.
Cited sources are review sites, publishers and marketplaces. They sell nothing in your category but compete with your pages for the citation slot.

Include generalist displacement: when an engine answers a niche prompt with a broad suite that covers your use case among many others, that brand takes the place your brand would have taken and belongs in the set.

The competitor set is also the brand list a tool must support, so check how many brands per project a plan allows before buying.

Prompts by buyer-journey stage, locked and dated

Our working rule, not a sourced benchmark, is to start with a couple of dozen prompts. Mix brand-neutral prompts, such as which tool suits a task, with brand-adjacent ones, such as alternatives to a named competitor.

Group prompts by stage: awareness, comparison and purchase. Report results per stage with a simple weighting you state next to the list, comparison and purchase prompts weighted above awareness prompts, and adjust it to your sales cycle. Losing a comparison prompt costs more than losing an awareness prompt, because the buyer is closer to a choice.

Lock and date the prompt set for the measurement period. If a prompt must change, version the set for the next period instead of editing it mid-period: editing changes the denominator of share of voice and breaks the comparison.

How many runs before a competitor gap is real

Run-to-run variance is the baseline, not a bug

LLM answers are non-deterministic: the same prompt on the same engine can name different brands across runs, so a single run measures noise. In research published on 28 January 2026, SparkToro reports less than a 1 in 100 chance that ChatGPT or Google's AI, asked 100 times, gives the same list of brands in any two responses.

Variance also differs by engine, partly because engines cite different numbers of sources per answer. Qwairy's citation study, published on 15 October 2025, reports that "Perplexity averages 21.87 citations per question." An engine that attaches many sources offers more citation slots than an engine that cites few. Measure each engine against its own baseline and never pool answers across engines.

Our working rule, not a sourced statistic: run each prompt several times per engine per period, spread across different days; count mention rate as the fraction of runs that name the brand; treat a gap no larger than a single run as a tie; and call a trend only when it holds in the following period too.

Use a clean browser profile or a temporary session without memory for every run. Memory and personalization can tilt answers toward brands you have discussed before.

Separating a real gain from normal fluctuation, yours or a competitor's

Apply the same persistence test to rivals. When a competitor's mention rate jumps, check whether every tracked brand rose on that engine: if so, the engine or model changed. If only the rival rose and the rise persists, look for a content or PR action that coincides with the jump, and date it from public evidence such as a publication date or press release.

A visibility score can move while your citations stay flat. Common causes: a changed denominator, with prompts or brands added; a new engine or model version; or a rival gaining mentions without citations, which lowers your share of voice without touching your citation count.

How to compare share of voice when every tool defines it differently

Some metrics survive any tool because you can recompute them by hand from archived answers: mention rate, citation rate, share of voice and average position. For the arithmetic, see how to calculate share of voice in AI search by hand.

Composite scores do not travel, because each tool sets its own engine list, prompt universe and aggregation. The Ahrefs Help Center, consulted on 5 October 2026, defines AI share of voice as "a brand's percentage share of impressions compared to other tracked brands." Semrush's knowledge base, also consulted on 5 October 2026, describes its AI Visibility Toolkit data as "a collection of over 317 million prompts and responses on ChatGPT, Gemini, Google AI Overviews, and AI Mode." Different units and engine sets cannot share a single axis.

No universal benchmark for a good AI visibility score exists. The meaningful benchmark is your relative share against named competitors on a locked prompt set.

Metric	Recomputable by hand	Comparable across tools	Best use
Mention rate	Yes, from archived answers	Yes, same prompts and engine	Presence per brand
Citation rate	Yes, from source lists	Yes, same prompts and engine	Which URLs win the source slot
Share of voice	Yes, over a fixed brand set	After normalization	Position against rivals
Average position	Yes, from answer order	Yes, same counting rule	Prominence inside answers
Proprietary composite score	No	No	Trends inside a single tool

To normalize, identify what the tool counts (mention, recommendation, impression or citation), map it to the matching raw metric, recompute share of voice over the same competitor set and engine, and compare only like with like.

When a brand leads on a given engine and trails on another, report each engine separately and prioritize the engine that matches your buyer's journey and market. An average hides the gap you need to act on.

A benchmark is valid for a single market: run the locked set per country and language instead of mixing locales.

Which way to benchmark: by hand, with DIY automation, or with a platform

The manual method is free. Use a clean browser profile, archive the answer from each run, and record mentions and citations per brand in a spreadsheet. Its limit is arithmetic: the workload is prompts times runs times engines, and every added engine or market multiplies it again.

DIY automation removes the copying. The terms of use of consumer AI products can restrict automated extraction of their output, so scripting a consumer interface is a risk to check product by product; official APIs are the compliant route. API answers and interface answers are not guaranteed to match, because defaults, retrieval and personalization differ, so hold the collection method constant for the whole benchmark and never mix methods. See how to automate AI visibility checks.

A monitoring platform should cover the engines your buyers use, allow enough competitor brands per project for your full set, and give access to raw counts and archived answers, not only a composite score.

Option	Cost	Effort per period	Engine coverage	Terms-of-use exposure	Auditability	Multi-brand support
By hand	Free	High, grows with prompts and engines	Any engine you can open	Low	Full if every answer is archived	Limited by your time
DIY via official APIs	API fees and build time	Low once built	Engines with a public API	Low on official APIs	Full, you store raw output	As many as you code
Monitoring platform	Subscription	Low	The vendor's engine list	Depends on the vendor's collection method	Depends on access to raw answers	Capped per plan

Build the scoring sheet with a row per run and these columns: date, engine, prompt id, journey stage, run, brands mentioned, position, URLs cited, description accurate (yes or no).

Add an accuracy check: when an answer describes a competitor or your brand incorrectly, log it as a separate finding. A wrong mention is a different problem from a missing mention and calls for a factual correction, not more coverage.

Where to start benchmarking against competitors this week

List the competitors. Pick a handful, spread across the co-mentioned, compared and cited groups.
Write the prompts. Draft a couple of dozen across awareness, comparison and purchase, then lock and date the list.
Run them. Repeat each prompt several times on more than a single engine, on different days, in a clean profile.
Record and compute. Log mentions, position, cited URLs and accuracy per run, then compute mention rate, citation rate and share of voice per brand, per engine and per stage.
Archive the baseline. Save and date the snapshot with its prompt-set version.

Once a gap holds, turn it into a prioritized list. Rank prompts by journey-stage weight multiplied by gap size. Open the competitor URLs cited on the top prompts and decide whether the fix is an owned page, a third-party source or a factual correction. Effort and impact here are judgement calls, not tested results. For patterns to look for, see the competitive insights AI visibility data reveals.

The benchmark starts to mean something at the second dated snapshot, when the same locked set runs again and each gap shows whether it holds.

Run a free Namedrop scan to see where your brand stands next to competitors in ChatGPT, Perplexity, Gemini and Claude, and keep it as the first dated snapshot of your benchmark.