CiteWeek · 2026-09-13 · Vendor prices re-verified 2026-09-13. Claims about our own method are labelled hypotheses not observed CiteWeek data.
When a buyer types "best [category] tool for [ICP]" into ChatGPT, does your
product come up? That question now has an entire category of dashboards
selling you the continuous version of the answer. The category anchor on our
radar is Vismore: $99/month at entry, $399/month at its Advanced tier
(vendor pricing pages, re-verified 2026-09-13; check again before you quote
it). Budget tiers sit far below that: Otterly's entry plan is $29/month
(re-verified 2026-09-13).
Here's the thing though: the first version of the answer costs you one
browser tab and half an hour. You should have that answer before you pay
anyone for the continuous version.
The 30-minute version
Open a fresh chat and ask the question the way a buyer would. Not
"generative engine optimization platform". For my product the buyer query is
"best weekly AI citation tracker for a small B2B SaaS team". If you sell to
accountants, ask like an accountant. If you sell to landscapers, ask like a
landscaper with two crews and a scheduling problem.
Then read the answer with two questions in mind:
- Were you named?
- If not, who or what was cited instead?
The second answer is the one people skip, and it's the more useful one. The
list the model produces instead of you is your real competitive set. In my
own category checks so far the named alternatives have been listicles and
comparison posts more often than the products my positioning page worries
about (hypothesis from a handful of informal runs, not a measured sample).
Write down date, query, engine, named Y/N, and cited-instead domain. Five
columns in a spreadsheet. That spreadsheet is the whole instrument. A $399
dashboard is, at its top line, a prettier and more continuous version of
those five columns plus some alerting.
One run is an anecdote
LLM answers are non-deterministic. Ask the same question twice and you can
get two different brand lists, which is exactly why "I asked once and I'm in
there" is not a finding. My working rule is 3-5 runs per query per week and
recording the majority answer. That's a hypothesis about a usable cadence,
not a validated method; the weekly cohorts we're running are what will tell
me whether mention frequency over N runs actually tracks anything a buyer
sees.
The weekly rhythm matters for a second reason. You can't change the model
directly. You publish something, wait, and re-probe. The feedback loop is a
week long no matter how often you check, so daily anxiety buys you nothing
that a weekly grid doesn't.
What the grid can't tell you
Being named is a binary, and binary metrics flatter you. "You're mentioned"
covers both "the cheap alternative to Y" and "the category standard". Only
one of those framings helps you, and no citation tracker I've seen scores
framing (if yours does, I want to see the methodology).
The grid also measures the logged-out, default-model answer. A buyer with
chat history and different settings can see something else entirely. Treat
your grid as the baseline case, not the buyer's exact experience.
And a good mention rate doesn't tell you what to do next. That comes from
reading the sources the answer leans on: which comparison pages, which
listicles, which threads. The fix for a weak result is almost always one
publishable artifact, like an honest comparison page or a tested-it-myself
post. That's a content task with a weekly cadence, not a dashboard
subscription.
If the result is bad
Bad result, meaning you're absent and a competitor is named: read the
sources behind the answer and publish the artifact they'd cite. A comparison
page is the highest-yield start because it's the artifact buyers' questions
map onto. Then re-probe next week and see if the answer moved. If it didn't
after a few honest attempts, you've learned something about how far content
alone gets you in your category, which is worth knowing before you rent a
$399 dashboard to watch it.
The free check
That five-column grid is exactly what CiteWeek runs for you weekly: three
buyer-language queries across ChatGPT, Perplexity and Google AI Overviews,
one email with named/cited-instead, and one article worth publishing. The
free check runs the grid once so you can see the shape of the data before
paying for anything. It's 3 checks per email per 14 days, no card, results
on the page:
When you repost or syndicate this, set utm_source to wherever it ran
(devto | medium | indiehackers | …) and keep utm_campaign as
article-1-does-chatgpt-recommend-you. That's how signups get attributed back
to this article in the funnel table, and I'd rather learn which articles work
from tagged links than from vibes.
Top comments (0)