A 20-minute monthly audit to see which sources AI assistants cite for your brand

Joseph Tannous — Fri, 16 Oct 2026 08:35:08 +0000

Marketers track rankings, traffic and share of voice. Few track the thing buyers increasingly see first: the short list an AI assistant gives when someone asks for a recommendation.

You can measure it in about 20 minutes a month. Here's the method we use, and what it showed when we ran it in our own category in October 2026.

Step 1: Write five real buyer questions (5 minutes)

Pick one of each:

A broad question ("What's the best [category] for [audience]?")
An industry question ("Best [category] for fintech companies")
A hyper-specific question (a named outcome, publication or integration)
A comparison ("[You] vs [competitor]")
A name check ("What is [your company] and is it any good?")

Step 2: Ask in a clean session (10 minutes)

Use a temporary chat with personalisation off and web search on, so past conversations don't skew the answer. After each answer, ask:

List the full URL of every page you used as a source.

Step 3: Log the sources by type (5 minutes)

One row per source: list article, vendor page, review site, news article, forum or docs. Note whether you appear in each.

Question	Source URL	Type	Are you in it?
Best X for Y	example.com/best-x	List article	No

What we found when we ran it

In October we ran this for B2B tech PR. For "best tech PR agency for enterprise software companies", ChatGPT used 15 sources, and 11 of them were third-party list articles. Broad questions were answered almost entirely from round-ups. Specific questions pulled in vendors' own pages that answered them directly. The name-check answer leaned on review counts and verifiable results.

It's a small, directional test (one run, one tool), which is exactly why the monthly log matters: the pattern only becomes trustworthy when you repeat it.

How to act on it

Lists in your sources: pitch the authors with something useful, such as data, a case study or an expert quote.
Competitors' pages cited for specific questions: publish a clearer, more specific page of your own.
A weak name-check: prioritise reviews and independent coverage.

Repeat monthly

AI answers shift as new content is published. A monthly log shows whether your work is moving the answer, which is the metric that increasingly matters.

Full method and the original results: https://www.jamcopr.com/ai-pr-agency-study

I'm Joseph Tannous, founder of Jamco PR.

I logged every source ChatGPT used to recommend a B2B vendor. Here's what it actually reads.

Joseph Tannous — Thu, 15 Oct 2026 13:32:11 +0000

If your product's buyers ask an AI assistant "what's the best tool/agency for X", what decides who gets named?

I run a PR agency, so I tested it in my own category. The method is simple enough to copy for yours.

The setup

Date: October 12, 2026
Tool: ChatGPT with web search on, in a temporary chat with personalisation off (no memory, no past conversations)
Method: ask a real buyer question, then ask: "List the full URL of every page you used as a source." Log everything.

What it relied on

Buyer question	Sources ChatGPT used
What is the best PR agency for B2B tech companies?	3: two "best of" list articles and a firm ranking, plus agency homepages
Best tech PR agency for enterprise software companies	15: 11 list articles, 1 agency blog post, 3 agency case studies
Which PR agency can guarantee my tech company gets published on Business Insider?	5: agency pages dedicated to that exact question, plus the publisher's own sponsored-content page
Best PR agency for SaaS companies that want top-tier coverage	5: sector pages and case studies with measurable results
"What is [our agency] and is it any good?"	Our site, our Clutch profile and our Trustpilot page

Three patterns

1. Broad questions are answered from lists, not from vendor sites. For the enterprise question, 11 of 15 sources were third-party "best X" articles. The vendors it recommended were the ones repeated across those lists. We weren't on them, so we weren't in the answer.

2. Specific questions reward specific pages. When the question named a publication and a guarantee, it stopped citing lists and cited vendors' own pages answering exactly that, especially pages explicit about terms ("what happens if it doesn't publish?").

3. Name-checks are judged on third-party proof. Asked about one vendor by name, it summarised the website accurately, but its verdict hinged on review counts and independently verifiable results. A small review count was enough for "do more due diligence".

How to run this for your own product

Write down 5 questions your buyers really ask (one broad, one by industry, one hyper-specific, one comparison, one about you by name).
Run each in a fresh temporary chat with search on.
Ask for every source URL and log them in a sheet: list article, vendor page, review site, docs, forum.
Repeat monthly. The source mix changes faster than Google rankings.

What to do with the results

Get onto the third-party lists your buyers' questions draw from.
Publish one specific page per real buyer question, with terms, numbers and proof.
Build reviews and independent coverage on purpose, because that's what the name-check reads.
Keep your company description consistent everywhere AI tools look.

Full method and results: https://www.jamcopr.com/ai-pr-agency-study

I'm Joseph Tannous, founder of Jamco PR. We rerun these questions monthly; follow for the trend data.

DEV Community: Joseph Tannous