AI Visibility Benchmark 2026: Share of Voice Report

Across six B2B software categories measured in August 2026, the category leader took a median 31.0% of all brand mentions in AI answers. The top two together took 46% to 65%. Share of voice varies by category and by engine, which is why a benchmark needs both dimensions.

The sample: 16,134 AI answers, 39,903 estimated brand mentions, 36 brands, six categories, six engines, United States, August 2026. Every share is calculated from Ahrefs Brand Radar's mention counts. Those counts are modelled estimates over a search-backed prompt corpus, not a census of what real people asked.

Key Takeaways

• The leader takes about a third: across six categories the top brand held 25.7% to 35.7% of brand mentions, median 31.0%. Nobody owns the answer.
• Sixth place is not zero: the weakest brand in each set still held 1.5% to 7.6%. Across these six-brand sets, sixth place still ranged from 1.5% to 7.6%, so a single-digit reading does not by itself indicate absence.
• Contrarian: high Domain Rating does not guarantee high share of voice. All brands here sit between DR 81 and 93, while their share of voice ranges from 1.5% to 35.7%.
• Engines agree more than they disagree: the median brand moved 4.2 points across six engines. The leading brand changed in two categories out of six.
• In the disclosed ChatGPT sentiment sample, 79.6% of 167 mentions assigned the brand to a customer segment, while 1.8% were framed primarily by a limitation.

Executive Summary

This ai visibility benchmark measures one thing. Of all the brand mentions an engine produces for a category's buying questions, it asks what proportion goes to each brand. The measure is relative, so it does not move with category popularity. That is what makes it board-ready. What visibility in AI search depends on covers the mechanism underneath it.

• Concentration varies more than leadership. The top two took 65.2% of CRM mentions and 46.1% of project management mentions.
• Answers name two to three brands, so the question is rarely winner-takes-all.
• Being cited beats being authoritative. Share of voice was more strongly associated with own-domain citations (Spearman rho +0.72) than with Domain Rating (+0.47) in the pooled sample.

Methodology and Sample

Categories Covered

Six categories entered the sample: CRM platforms, accounting software, payroll and HR, project management, web hosting, and email marketing. A category ships only if its scoped set clears 900 answers. Four were measured and cut on that rule. Endpoint security returned 130 answers, help desk software 482, ecommerce platforms 625, and e-signature 625.

Brands per Category

Six brands per category, 36 in total, chosen as the names a buyer would shortlist. The brand set defines the denominator, so share of voice means share among these six.

How an Answer Enters a Category Set

An answer counts when the question carries one of the category's terms and names at least one of the six brands. The questions are Ahrefs' search-backed prompts. They are built from Google's People Also Ask corpus and semantic fan-out, then run through each engine's public web interface. The term list is the method, so it is published in full.

Scoping by question is the finding underneath the finding. Ahrefs matches brand mentions as strings. Counting a name wherever it appears inflates any brand named with an ordinary English word. Only 2.5% of Workday's mentions sat inside payroll questions. Intercom's ran at 1.3%. Salesforce ran at 17.7% and ADP at 26.8%. The rest were answers about a work day and about door intercoms.

Run Cadence

One pull, United States, August 2026, six engines: ChatGPT, Microsoft Copilot, Google Gemini, Perplexity, Google AI Overviews, and Google AI Mode. The four chatbot indexes refresh monthly on a rolling 90-day window. The two Google surfaces refresh continuously. A figure here describes a window, not a day.

Two engines could not be measured. Ahrefs tracks Claude through custom prompts only, so it sits outside the search-backed index used here. Grok is in that index. Ahrefs records that it is temporarily unable to gather new Grok data after a policy change at Grok. Every Grok query returned zero.

Edition naming. Each edition is named for its publication year and reports the most recent complete window. The 2027 edition therefore carries an August 2026 pull. Klarivo sells AI visibility monitoring. Read the leaders and laggards section with that interest in view.

Share of Voice by Category

Category share of voice is each brand's estimated mentions divided by all six brands' mentions in the same scoped set. Sample sizes sit in the table because a benchmark without a visible n is marketing.

A third is the leader's benchmark, not a ceiling. A brand at 12% is not failing. It is mid-table.

Concentration is the variable to check first. CRM and accounting are top-heavy, the leading pair taking six mentions in ten. Project management spreads across four brands inside eight points. The same 15% share means second place in one category and fourth in another.

Engine-by-Engine Variance

Engines were measured separately on the same scoped sets. The table reports the pooled leader's share on each engine.

Cell sizes ran from 101 answers (project management on AI Overviews) to 1,617 (CRM on Copilot). No cell below 100 was published.

Across all 36 brands the median engine spread was 4.2 points, with a maximum of 12.5. The broad competitive picture was often similar, but individual engines could still produce materially different rankings.

Where they disagree, they disagree about first place. The leading brand changed by engine in two of six categories. Salesforce led CRM on ChatGPT, Copilot, and AI Mode. HubSpot led on the other three, on a gap under two points. ADP led payroll on three engines and Gusto on the rest. A single-engine reading hands you a leadership claim your rival disproves on the next engine.

Sentiment Benchmarks

Sentiment has to be read at the sentence naming the brand, not at the answer, so it was scored on a disclosed sample. Sample: the five most relevant ChatGPT answers per category, 30 answers, 167 brand mentions, August 2026. Each mention was scored once, against four bands.

Counts by category, because 24 to 30 mentions is too thin to carry a percentage.

Negative framing barely exists. Three mentions in 167. A separate count found 20 mentions (12.0%) carrying an explicit drawback clause, such as "can become expensive as teams grow". Those sat mostly in comparison-table answers. Reputation damage is not the common failure mode.

Segment assignment is the real contest. Four brands in five are handed a customer type rather than a verdict. The recommendation slot is scarce and not always stable. QuickBooks took it in five of five sampled accounting answers; the email marketing slot went to three different brands.

What Separates Leaders From Laggards

Two explanations were tested against the same 36 brands: Domain Rating, and own-domain citations inside the category's answer set. Spearman rank correlation, six brands per category.

What Leaders Have in Common

Leaders are cited on their own domain inside their category's answers. Salesforce's domain appeared in 1,436 CRM answers, QuickBooks in 1,216, ADP in 807. Each leads its category. In this sample, higher share of voice was positively associated with more own-domain citations. Publishing so machines can consume it is the discipline underneath that.

What Laggards Share

Several laggards combined low share of voice with relatively few own-domain citations, while the sentiment sample provided little evidence of strong negative framing. HostGator held 2.8% of web hosting mentions with 30 own-domain citations. Freshworks held 1.5% of CRM mentions. Nothing in the sampled answers argued against either. They were not in the material the engines drew on.

What Made No Difference

Domain Rating did not separate this field. All 36 brands rate 81 to 93, median 90, while share of voice ranges from 1.5% to 35.7%. Two counterexamples sharpen the point. Trello held 20.4% of project management mentions with its own domain cited in 12 answers. Wrike held 7.6% with 195. Third parties carry Trello; Wrike's own pages are read without being recommended.

One caution. Own-domain citations and mentions are not independent, because an answer citing your domain often names you. Treat +0.72 as how the two measures move together, not as proof that publishing more pages raises share of voice.

Limitations

Six brands per category is a small field. Rank correlations on six points are indicative only. The pooled figure across 35 brands is the one to quote.

Two engines are missing for different reasons. Claude sits outside the search-backed index, and Grok collection is paused at the source. A team tracking five engines internally will not reconcile line by line with this table.

The prompts are modelled, not observed. They expand Google People Also Ask questions rather than sampling what buyers typed into a chatbot. Locale follows the source keyword, so "United States" describes the query, not the audience.

Four categories were measured and cut for thin samples: endpoint security, help desk software, ecommerce platforms, and e-signature.

Ordinary-word brand names distort any name-matched measure. Question scoping removes most of it, not all. Workday should expect a wider error bar than Pipedrive.

The sentiment sample is 167 mentions on one engine. Large enough to describe how brands are framed, too small to score a single brand.

How Klarivo Benchmarks a Category

A benchmark like this one is a snapshot. The version that changes decisions is the same method run continuously on your category, your competitor set, and your buying questions. That means fixing the brand list and the question set, then holding both still long enough that a movement means something.

Klarivo Monitor runs that measurement across five answer engines: OpenAI's ChatGPT, Anthropic's Claude, Perplexity, Google Gemini, and xAI's Grok. Each is tracked independently and refreshed on a schedule the client controls. It reports competitor share of voice alongside query-level mention rates and sentiment trends. Two of those five are the engines this study could not reach.

The hardest finding here to reproduce alone is the leaders and laggards test, because it needs the citation side and the mention side measured together. Watching competitor share of voice across engines tells you where you stand, and the domains the engines drew on tell you what to do next.

To get your own category measured this way, fifteen minutes is the whole commitment. Choose a slot, get instant confirmation, and fill in no form first. Book a Klarivo discovery call.

Download the Dataset

There is no file to fetch and no form to fill in. The dataset is the table below, printed in full so it can be copied, quoted, or pasted into a sheet. Own-domain citations count answers in the category's set citing the brand's domain, subdomains included. QuickBooks combines quickbooks.intuit.com and intuit.com. Zoho CRM and Zoho Books share zoho.com, counted inside each category's own set.

Microsoft Dynamics 365 carries no citation or Domain Rating figure because its pages sit on the shared microsoft.com estate. It stays in the share of voice column and out of the correlations.

To rebuild it from scratch: Ahrefs Brand Radar, mentions overview by entity, country US. Data sources chatgpt, copilot, gemini, perplexity, google_ai_overviews and google_ai_mode. One brand set per category, with a question filter built from the category terms table above. Those parameters regenerate the underlying Brand Radar mention and citation data used in the share-of-voice and engine-variance analysis. The sentiment and correlation sections additionally require the scoring and calculations described in their respective methods.

How to Cite

Preferred citation: Klarivo, AI Visibility Benchmark 2026: Share of Voice Report, https://www.klarivo.ai/blogs/ai-visibility-benchmark, August 2026.

When quoting a figure, carry the sample: "the category leader took a median 31.0% of brand mentions across six B2B software categories (Klarivo, AI Visibility Benchmark, n = 16,134 AI answers, United States, August 2026)."

Underlying data source: Ahrefs, Brand Radar Methodology: How We Collect and Model AI Visibility Data, updated February 2026, and the Ahrefs Help Centre entry What is Brand Radar, July 2026.

Reuse terms: reproduce the tables in full with attribution and a link.