Skip to content

A model answers from two memories. LLM SEO services should work on both.

What it learned in training, and what it looks up live when it searches. ChatGPT, Gemini, Perplexity and Google's AI answers lean on each in different measure, and each memory moves on its own clock.

Most LLM SEO services work on the live half, because it reports fastest. I work on both, and the audit shows which one is getting you wrong.

  • 2

    memories, two clocks

  • 5

    AI engines checked

  • 150

    buyer prompts, weekly

  • 2 weeks

    to the full audit

Search and AI answer work for

Founders and marketing teams, B2B and B2C.

  • Procured
  • Procured

Two memories. Each gets you wrong in its own way.

ChatGPT optimization services usually pick one memory and call it the whole job. The symptoms tell you which one is the problem.

Memory one

What the model learned

Everything it read before its knowledge cutoff, compressed into the model. It answers from this when it does not search.

You see it when

  • It describes your old pricing, or a product you retired.
  • It names your competitors and not you, with search switched off.
What moves it
Being described the same way on many independent sites, long enough for the next model to read it.
Its clock
Slow, and set by the vendor: the next model version.

Memory two

What the model looks up

When an engine searches, it runs its own queries, reads a handful of pages and cites some of them in the answer.

You see it when

  • A rival's comparison page is cited for a question you should own.
  • Your page ranks on Google, and the engine still never quotes it.
What moves it
Pages that rank for the searches the engine runs, readable by its crawler, with the answer in one quotable sentence.
Its clock
Fast: once a page is crawled, indexed and retrieved.

Three neighbors of this page take one slice each: how a passage gets quoted, the search index under the live half and measuring answers that have no rank. This one is about which memory you are fixing, and how long it takes.

Pick an engine. See which memory answers.

Five situations your buyers are in every day. Each leans on a different memory, so each needs different work.

Training memoryAnswers

What the model learned before its cutoff. Changes only with a new model.

Live retrievalIdle

The searches it runs now, the pages it reads, the sources it cites.

What answers
The model's own memory. No pages are fetched and nothing is cited, so the answer reflects what it read before its cutoff.
What moves it
A consistent description of you across many independent sites, in place before the next model is trained.
How fast
The slowest clock here. It changes when a new model ships, which nobody outside the vendor schedules.
What the audit checks
Your brand and category prompts asked with search off: wrong facts, missing products, rivals named instead of you.

How each engine behaves as of November 2026. Engines change how and when they search, so the audit checks again rather than trusting this summary.

Three companies, three numbers you can check.

Each moved first on the live half, the clock you can move this quarter: pages the engines could find, read and quote. Each links to the full case study.

  • VoiceOrder

    9 of 10

    monitored AI prompts citing the site, after month one

    • From absent to cited in one month
    • 15 articles written for the questions buyers ask
    • 3 links and 1 mention on sites the engines already cite
    • Technical fixes on Webflow: crawl, metadata, canonicals
  • Awesomic

    #1

    most-named of 12 competitors in AI answers, in 3 months

    • 30% share of voice across 107 tracked prompts
    • 4th most-cited website in its category, behind only Google, YouTube and Reddit
    • 200 to 300 sessions a week from AI search
    • Built from 50+ listicles, 100+ use-case pages and 20+ outreach links a month
  • Strada

    7×

    daily organic visitors, 10 to 70, in 6 months

    • 35% average month-over-month growth
    • About 500 sessions from AI answers, with no listicles or third-party mentions
    • 8 to 12 pieces a month written for enterprise insurance buyers
    • 20+ backlinks a month from insurance tech and adjacent SaaS sites

Want to know which memory is getting you wrong?

The audit is $2,500 and takes two weeks. Both memories checked on 5 engines, a 90-day plan and every fix, and you keep all of it even if we never work together.

Two weeks, $2,500. Both memories, checked and planned.

Specific enough for your team, a freelancer or any other provider to run. Yours to keep, even if you never hire me.

  • 150 × 5

    Your baseline in AI answers

    150 buyer prompts on ChatGPT, Perplexity, Gemini, Google AI Mode and AI Overviews: who gets named, and which sources each answer was built from.

  • Search off

    What the models say from memory

    Your brand and category asked with web search off: wrong facts, retired products and the rivals named instead.

  • Category

    The full category analysis

    Every competitor that matters: what they rank for, what the AI answers say about them, and where your gap is cheapest to close.

  • 90 days

    A 90-day plan, in order

    What to publish, update or retire for the next 90 days, each item tied to a query or prompt and a buyer stage, sequenced by what should move first.

  • Targets

    Link and mention targets

    The sites that already cover your category, who runs them, and the angle to pitch. Earned through outreach, never bought or sold.

  • Fixes

    Every technical fix

    Crawl, indexing, rendering, schema and AI crawler access, as a prioritized list your developer can ship.

Book the audit

Fixed price, two weeks, one call to start.

What LLM SEO services cannot promise, whoever sells them.

Anyone selling this work has the same two memories to work with and controls neither. Four limits hold for every provider, including this one.

  1. 01

    Nobody can put you into a model's training data.

    What goes into a training set, and when, is the vendor's decision. The work makes you easy to learn about. Anyone promising inclusion is selling a guess.

  2. 02

    A wrong fact can outlive the fix.

    Correct your pricing today and the answers that search can update within weeks. The model's own memory keeps the old version until a newer model replaces it.

  3. 03

    Retrieval is only as good as the search underneath it.

    Engines that search pull from an index. If your pages are not indexed, or do not rank for the searches the engine runs, no amount of prompt work reaches them.

  4. 04

    Nobody can see inside the model.

    There is no console showing what a model knows about you. Measurement is sampling: the same prompts, asked again and again, read for patterns rather than positions.

Each limit is a reason to measure the two memories separately. That is how the plan gets made, and how you will know whether it worked.

Six workstreams, split by memory.

The retrieval work ships first because it moves first. The training work runs alongside, because the next model is already reading.

  1. 01

    Weeks 1 to 2

    Fix the facts at the source

    Your site, docs, pricing page and profiles say the same thing in the same words. Both memories start from these pages.

  2. 02

    Week 1

    Open the right doors

    Robots rules for AI crawlers, set one by one: training crawlers and search crawlers are separate agents with separate switches.

  3. 03

    From week 2

    Rank for the searches the engines run

    Engines rewrite a buyer's question into their own searches. Pages are planned against those searches, not only the question.

  4. 04

    Ongoing

    Earn the record the next model reads

    Mentions and links on independent sites your category already trusts, earned through outreach. Never bought or sold.

  5. 05

    When found

    Correct what is wrong

    When an answer repeats an outdated fact, trace the page it came from and get that page updated, yours or someone else's.

  6. 06

    Weekly and monthly

    Measure both memories

    150 prompts on 5 engines every week for what they look up. A search-off check each month for what they remember.

If most of your gap is the quoting itself, start with the passage-level work.

What clients say about working with me.

Founders and marketing leads I have worked with on search and AI answers. Real names, real companies.

  • “Eugene helped with our marketing efforts at Strada and has been terrific to work with. He's professional, dependable, and always finds a way to deliver high-quality work.”

    Amir Prodensky

    Amir Prodensky

    CEO / Strada

  • “Working with Yevhen is an absolute pleasure. His marketing analysis is some of the most in-depth I've ever seen. He consistently goes above and beyond expectations.”

    Ira Nykytenko

    Ira Nykytenko

    Operations Manager / Awesomic

  • “Yevhen consistently delivered content on time, strictly adhering to briefs while frequently exceeding expectations. What sets Yevhen apart is balancing technical SEO with engaging, reader-friendly content.”

    Konstantin Vashkevich

    Konstantin Vashkevich

    Head of Marketing / RedTrack.io

  • “What truly sets Eugene apart is his deep understanding of SaaS content marketing. He knows how to make content work hard – pulling in the right audience and converting them.”

    Andrew Kolp

    Andrew Kolp

    Founder / Generect

  • “His contributions have been incredibly valuable to our blog. I have immense confidence in his knowledge and expertise. He is undoubtedly a professional in content.”

    Zeynep Kantav

    Zeynep Kantav

    Marketing Manager / Transkriptor

  • “Eugene is one of those rare professionals who combines incredible skills with an unbeatable work ethic and amazing attitude. He creates experiences that grab attention and drive results.”

    Rimma Sytnik

    Rimma Sytnik

    Head of Inbound Marketing / Reply

Four ways to buy this, and which memory each one covers.

LLM optimization services come in three common shapes: a specialist provider, a prompt-tracking tool, or a content shop that writes for AI answers. Here is what each one covers, and what it leaves to you.

busyless
LLM SEO agency
Prompt-tracking tool
Content shop
What you buy
One operator who owns both memories and the search work under them
Usually a retrieval program, reported monthly
A dashboard of prompts. The fixing is yours
Articles shaped for AI answers
Training memory
In scope: the consistent record and the outreach that builds it
Sometimes, sold as brand work
Not visible: tools read live answers
Not in scope
Live retrieval
In scope: crawler access, pages for the engines' searches, quotable answers
In scope
Measured, not acted on
The writing, without the ranking work
Measurement
150 prompts on 5 engines weekly, plus a monthly search-off check
Vendor dashboard, resold
Whatever the tool tracks
Word counts and publish dates
Who does the thinking
The person you met on the call
A strategist, then juniors you never meet
You
A writer, briefed by you
Cost shape
$2,500 audit, then from $5,000/mo to run it
Monthly retainer, usually annual
$100 to $1,000+/mo, plus your time
Per article

If you only need the tracking, a tool is the cheaper buy. If you have a strategist and need words, a content shop fills that gap. This is for teams that need one person to own both memories.

Find out which memory is wrong first. Then fix it.

Most LLM SEO services quote a retainer before anyone has asked the engines what they say about you. The audit asks first, with search on and off, so the scope comes from the answers.

Start here

Audit

$2,500one-off · two weeks

Two weeks. Both memories checked on 5 engines, the category analysis and a 90-day plan, specific enough for anyone to execute. You keep all of it whether or not we carry on.

  • 150 prompts on 5 engines: 750 answers read
  • What the models say from memory, with search off
  • 90-day plan, link and mention targets
  • Every technical fix, crawler access included
Book the diagnosis

Most common

Ownership

From $5,000per month

All six workstreams, both memories, measured weekly. Production included. Month to month, because a provider who needs a year-long lock-in is telling you something about their confidence.

  • Retrieval work first, training work alongside
  • Weekly runs on 150 prompts, one read a month
  • First content live in about 14 days
  • Run inside the wider content strategy, not beside it
Talk it through

Production is included, in two package sizes, on the main pricing page.

One call. Real plan, not a pitch.

30 minutes. We talk about what's already working, who owns content today, and whether a fractional Head of Content is actually the right move. If it isn't, I'll say so. Before the call I look at your site and bring three quick fixes you can ship either way.

Direct calendar

Book a 30-min intro call

Loading scheduler…

Calendar busy?

Send a note instead.

One sentence on the bottleneck. I'll reply within 24h with a sharper next step.

Or send a note

Questions, ahead of time.

  • What is LLM SEO, and how is it different from ordinary SEO?+
    LLM SEO is the work of being described correctly, and cited, by large language models: in what they remember and in what they look up. Ordinary SEO wins a place in a list of links. Some call it ChatGPT SEO, but the same work covers Gemini, Perplexity and Google's AI answers.
  • Can LLM SEO change what ChatGPT says about us when it does not search?+
    Only slowly. Without search, ChatGPT answers from what its model learned, and that changes when a new model is trained and released. The work makes the record about you consistent and widely corroborated before that happens. Nobody can say which model picks it up, or when.
  • How fast does LLM SEO show results?+
    The retrieval half can move in weeks: VoiceOrder went from absent to cited in 9 of 10 monitored prompts after month one. The training half moves with model releases, which nobody outside the vendors schedules. Month three is the first read worth a budget decision.
  • Should we block AI training bots, and does that hurt LLM SEO?+
    It is a trade. Training and search crawlers are separate agents, so you can block one and allow the other. Blocking a search crawler takes you out of that engine's cited sources. Blocking a training crawler changes what future models learn from your own pages, not what others write about you.
  • Do LLM SEO results show up in our analytics?+
    Partly. Answers that cite you can send referral visits you can see. Answers from memory send nothing, and many answers end the search without a click. So the read pairs weekly prompt tracking with branded search and assisted conversions, and says where attribution stops.
  • Is LLM SEO the same as AEO or GEO?+
    Same family, different emphasis. LLM SEO, as this page uses it, is about the two memories and their clocks. AEO is usually sold on measurement, GEO on how a passage gets quoted. One audit covers all three, because the engines do not care what the work is called.
  • What does LLM SEO cost here, and what comes first?+
    $2,500 for the two-week audit: 150 prompts on 5 engines, a search-off check of what the models remember, the category analysis, a 90-day plan, link and mention targets and every technical fix. Running it starts at $5,000 a month, production included, month to month.
  • Do you buy mentions or links for LLM SEO?+
    No. Mentions and links come from outreach to sites that already cover your category: somebody choosing to include you. Nothing is bought or sold. A paid placement is a weak signal, and a record built on them does not hold up when the next model reads it.