Published measurement, September 2026

AI Answer Benchmark

Who the AI engines actually name when a buyer asks for help with AI search, and what they read to decide

This is one week of measurement, published whole: the questions asked, the platforms they were asked on, the date, the total number of answers, every firm the answers named with its count, and every source the answers cited. AI Search Engine Optimization is in the table at its real position, near the bottom.

It is published because most AI visibility claims cannot be checked. A review of 27 firms offering AI visibility work, cited by ChatGPT in one of the answers captured below, found that 22 of them did not disclose the prompt set, the measurement period or the denominator behind their numbers. Everything needed to reproduce this page is on it.

30 minutes with Phil. Your questions, your competitors, your baseline.

247 mentionsCounted across the week

18 firms namedBy the answers themselves

15 sourcesCited more than once

The method, in full

How this was measured

Read this before the numbers. A count means nothing without the conditions that produced it.

What was asked

26 questions, split between buyer questions where being named is the result (for example, the best AI SEO agency, the best AI SEO company for small businesses, who can audit a website for AI search visibility) and research questions where being cited is the result (for example, what entity SEO is, how to run an internal link audit, how to fix keyword cannibalization).

Why it matters: a firm can look invisible on one set and strong on the other, so the two are counted separately and never averaged into one score.

Where it was asked

Six platforms: ChatGPT, ChatGPT with web search, Google AI Overviews, Google AI Mode, Gemini and Bing Copilot. Each question was put to each platform independently, in the United States.

Why it matters: the platforms disagree. In this run the same buyer question was answered with agency names by ChatGPT and Google AI Overviews, and with a table of software tools by Google AI Mode and Bing Copilot.

When, and how often

The seven days to September 16, 2026. The same set runs every week, so each figure here has a previous value and will have a next one.

Why it matters: single answers move. On September 13 one question named this firm first of five; on September 16 the same question named three larger firms and not this one. A benchmark is a series, not a screenshot.

The denominator

247 brand mentions were counted across the answers to those 26 questions in the window. The percentages in the source table are the share of captured answers in which a domain appeared at least once.

Why it matters: a share of voice with no denominator cannot be checked. Answers that returned no source list are recorded as unmeasured, never as zero.

How a firm is detected

Each firm is matched by name, by its known aliases and by its domain. Firms the answers named that were not on the starting list were added rather than ignored; five were added this way in this run.

Why it matters: a detection rule that only looks for the firms you expected returns a flattering, wrong answer.

What is not claimed

Model outputs vary between runs and between users, and answers are localized in ways that are not always disclosed. In this run one question localized to three different cities on successive attempts, adding different local firms each time.

Why it matters: nobody can guarantee a particular AI answer. Anyone promising one is describing something they cannot control.

Result one

Every firm the answers named

Each firm the AI answers named in the seven days to September 16, 2026, with the number of answers naming it and its average position in the lists it appeared in. Lower average position is better; it is read together with the count, never instead of it.

# Firm named by the answers Answers naming it Average position
1 iPullRank 53 2.9
2 Siege Media 42 4.3
3 First Page Sage 35 4.4
4 Victorious 24 5.5
5 Searchbloom 19 5.3
6 NoGood 15 3.9
7 Directive Consulting 14 5.4
8 Animalz 14 6.1
9 Percepture 12 3.1
10 SeoProfy 5 5.0
11 Complete SEO 3 2.0
12 Graphite 2 7.0
13 GreenBanana SEO 2 8.8
14 Single Grain 2 11.0
15 AI Search Engine Optimization 2 1.5
16 Marcel Digital 1 1.0
17 Obility 1 1.0
18 Amsive 1 4.0

Source: weekly measurement by AI Search Engine Optimization, 26 questions across six AI platforms, United States, seven days to September 16, 2026. Firms are listed because AI answers named them; inclusion is not an endorsement and the table is not a ranking of quality.

Result two

What the answers read before they answered

The sources cited most often across the same answers, with the share of captured answers each appeared in. This is the more useful table of the two, because it says where a recommendation is decided.

Source What it is Share of answers
developers.google.com Google’s own developer and Search documentation 21.7%
youtube.com Video 16.7%
semrush.com SEO tool vendor 14.4%
ahrefs.com SEO tool vendor 13.2%
clutch.co Review directory 9.7%
screamingfrog.co.uk SEO tool vendor 8.0%
reddit.com Community threads 8.0%
firstpagesage.com Agency’s own published lists 7.3%
searchengineland.com Trade publication 6.4%
thriveagency.com Agency’s own published guides 5.0%
loudface.co Third-party agency list 4.5%
vogue.com Consumer magazine 4.5%
yesoptimist.com Third-party agency list 4.5%
ayrank.com Third-party agency list 4.5%
pikaseo.com Third-party agency list 4.0%

Same measurement and window as the table above.

Result three

Four things this week’s data says

Read from the two tables above, with nothing added.

The top three firms hold most of the ground

iPullRank, Siege Media and First Page Sage were named in 53, 42 and 35 answers. The next fifteen firms together account for fewer mentions than those three.

What follows: in a market this concentrated, moving from unnamed to named is a placement problem before it is a website problem.

No agency website is a leading source

The most-cited sources are Google’s own documentation and YouTube, then SEO tool vendors, a review directory and Reddit. The two agency domains that appear are cited for their published lists and guides, not their service pages.

What follows: an agency that only improves its own service pages is competing for a slot the answers rarely give to anyone.

Video is a citation surface, not a branding exercise

YouTube appeared in 16.7 percent of answers, ahead of every trade publication in the set and ahead of every agency site.

What follows: for research questions on Google’s AI surfaces, a clear video answer competes with written pages rather than supporting them.

Position without volume flatters

AI Search Engine Optimization averaged position 1.5 across the two answers that named it. The most-named firm averaged 2.9 across 53.

What follows: any agency reporting only average position, this one included, is showing you the better half of its number. Both belong in a report.

Using this

What to do
with a benchmark

A benchmark for your own market is the first deliverable of an AI visibility audit, and the measure everything afterwards is judged against. The terms used here are defined, with dated findings, in the AI search glossary; the measurements to expect from anyone you hire are set out on the AI SEO services page.

This page is updated as the measurement moves, and every figure carries the date of the run it came from. When a number here changes, the previous value stays visible in the update note, so the direction can be read rather than taken on trust.

Your next move

See this table
for your market

Book a free 30-minute strategy call with Phil. Bring your market and your competitors; leave knowing which questions to measure, who the AI names instead of you today, and which source is deciding it.

Book your free strategy callNo obligation. A useful next step.