Open data
AI Search Visibility Benchmark Dataset
Which brands do large language models actually name when people ask about SEO tools? Here is the complete underlying data — 100 queries, 3,117 rows, nothing behind a signup.
We ran 100 real SEO-category queries through an aggregated corpus of language-model responses and counted every brand cited. Then we pulled Google’s organic top 10 for the same 100 queries, so the two channels can be compared directly rather than assumed to agree.
This page is the data itself. The article interprets it; these five files let you disagree with the interpretation.
Read the analysis: AI Search Visibility Benchmark.
The headline numbers
- 100 — queries, across seven intent blocks
- 11,654 — raw brand mentions counted
- 271 — distinct brands after cleaning
- 947 — Google organic results, same queries
- 56% — of tool mentions go to just three brands
- 0 — mentions for AnalySEO, who published this
That last one is not modesty. A benchmark where the author comes first is not a benchmark, and the second edition is only interesting if the first one is honest.
The top ten, as the data has it
Mentions summed across the 100 queries, SEO tools only. Website platforms are counted separately and sit in the file, not in this chart.
| # | Brand | Mentions | Queries |
|---|---|---|---|
| 1 | Semrush | 1,201 | 78/100 |
| 2 | Ahrefs | 1,155 | 78/100 |
| 3 | Google Search Console | 1,035 | 72/100 |
| 4 | Google Analytics | 325 | 40/100 |
| 5 | Moz | 297 | 48/100 |
| 6 | Google Keyword Planner | 274 | 45/100 |
| 7 | Ubersuggest | 159 | 36/100 |
| 8 | SE Ranking | 153 | 32/100 |
| 9 | Yoast | 128 | 18/100 |
| 10 | Screaming Frog | 81 | 19/100 |
Three of the top six are free Google products. Together they account for 1,634 mentions — more than Semrush or Ahrefs on their own.
The five files
Every file joins on query_id. UTF-8, comma-separated, quoted where needed — they open in Excel, Sheets and pandas without any coaxing.
| File | Rows | Size | What it holds |
|---|---|---|---|
| 00-queries.csv | 100 | 4 KB | The query list with its intent block. The unit of analysis. |
| 01-brand-ranking.csv | 271 | 13 KB | Every brand, cleaned and merged, with mentions broken down by block. |
| 02-mentions-by-query.csv | 1,763 | 122 KB | Every mention, query by query, before filtering. The one that matters. |
| 03-google-serp.csv | 947 | 106 KB | Google’s organic top 10 for the same 100 queries. |
| 04-brand-domains.csv | 36 | 1 KB | The brand-to-domain map, so the Google overlap number is reproducible. |
The unfiltered file is the important one. Every brand we removed as a generic descriptor is still in there, flagged under filtered_as_generic, so you can redo the cleaning differently and see whether the conclusions survive. We think they do. Check.
Raw API responses: benchmark-ia-menciones.json and benchmark-serp-google.json.
What the columns mean
The ranking file carries most of the analysis. The rest is self-describing.
| Column | Meaning |
|---|---|
| rank_overall | Position across everything, tools and platforms together |
| rank_within_type | Position among SEO tools only, or among platforms only |
| brand | Canonical name, after merging case and naming variants |
| type | seo_tool, or platform for website builders and ecommerce platforms |
| total_mentions | Mentions summed across all 100 queries |
| queries_present | How many of the 100 queries the brand appears in |
| discovery … price_free | The same total, split across the seven intent blocks |
| 02 · filtered_as_generic | yes means we excluded it from the ranking as a descriptor, not a brand. Nothing is deleted. |
| 03 · domain_type | tool_vendor is a tool’s own site; publisher is somebody writing about tools. |
Method
Source
Brand mentions come from DataForSEO’s LLM mentions corpus, aggregated across model responses rather than collected from single live prompts. That matters: aggregating over a corpus avoids the run-to-run non-determinism that makes one-shot prompting unreliable. Organic results come from DataForSEO’s Google SERP API. Market: United States, English. Collected September 2026. Total API cost: $12.10.
Query selection
100 queries across seven blocks: discovery (18), alternatives (16), tasks (22), comparisons (12), segments (12), AEO/GEO (10), price and free (10). Weighted by intent rather than by search volume, because a citation benchmark measures the contexts in which brands get named, not demand. No query contains a year, so the same list is rerunnable and comparable quarter to quarter.
Cleaning, pass 1 — case variants
The API counted SEMrush and Semrush separately, at 647 and 554. Merged, Semrush moves from fifth place to first. The same applied to Moz and Moz Pro, Surfer and Surfer SEO, KWFinder and Mangools.
Cleaning, pass 2 — generic descriptors
The API returns phrases as if they were brands. The phrase “SEO software company” accumulated 267 mentions and would have placed twelfth. Removing generics dropped 9% of raw mentions, and every removal stays visible in file 02.
Cleaning, pass 3 — platforms
Wix, WordPress, Shopify and similar are tagged as platforms rather than deleted. They are not SEO tools, but Wix alone was mentioned more often than Moz, Ubersuggest and SE Ranking combined, so hiding them would misrepresent the data.
The counts
444 distinct brand strings in the raw response, 437 once case variants are merged, 11,654 raw mentions. After cleaning: 271 brands and 9,706 mentions, of which 255 brands are SEO tools. 103 of those were mentioned exactly once. Six of the 100 queries returned no brands at all, and are reported as zero rather than replaced.
What this dataset does not claim
- It does not measure market share, revenue or product quality. It measures how often a brand is named.
- It is not a sample of all AI. It is one corpus, with its own coverage and biases, and it is not independently reproducible by a third party. That is a real limitation and we would rather state it than have it pointed out.
- 100 queries cover the category. They do not exhaust it.
- It is a single point in time. The value arrives with the second edition, when there is movement to report.
Licence and citation
Published under CC BY 4.0. Use it, chart it, quote it, disagree with it in public. The only condition is a link back. Read the licence.
AnalySEO (2026). AI Search Visibility Benchmark, Edition 1. https://analyseo.app/data/ai-visibility-benchmark
Found an error? Write to hello@analyseo.app. Corrections go into the next edition with a note saying what changed and who caught it.
Frequently asked questions
Which models does this cover?
It is an aggregated corpus of language-model responses supplied by DataForSEO, not a single named model. That is a genuine limitation: you cannot use this to say that one particular assistant prefers a given brand. What you can say is which brands dominate model-generated answers about SEO tools in general, which is the question most people are actually asking.
Why not just prompt an assistant 100 times yourself?
Because the same prompt returns different brands on different runs. Any single-shot benchmark measures the sampling temperature as much as the brand landscape. Aggregating over a corpus is the less exciting method and the more defensible one.
Why is AnalySEO not in the ranking?
Because it was mentioned zero times. So were DinoRank, SEOptimer, SearchAtlas and Sistrix. Publishing that is the point: it is the baseline we will be measured against in the second edition.
Can I use this commercially?
Yes. CC BY 4.0 permits commercial use, including analysis you sell. Attribute the source and link back.
How often will it be updated?
Quarterly, with the same 100 queries and the same method, so the editions are comparable. Changing the query list would make the comparison meaningless, so the list is frozen.
Can I get the raw API responses?
They are published alongside the CSVs as JSON, in case you want to redo the cleaning from scratch. You probably should, and if your cleaning changes a conclusion we want to hear about it.