How We Get Clients Featured on ChatGPT and Claude: 5 Growth Hacks

You cannot buy a slot inside an AI answer. That is exactly why it is the most defensible marketing real estate in crypto.

By Yuval Halevi, GuerrillaBuzz · Published June 2026 · 11 min read

From my experience, the question every founder asks in 2026 is no longer "how do we rank." It is "when someone asks ChatGPT who to use, why is it never us." This piece is the answer: the five plays we run at GuerrillaBuzz to put companies inside that response.

One framing before the plays. There is no ad unit inside ChatGPT or Claude. No bidding, no rate card. Every recommendation is assembled from sources the model trusts, which means every recommendation can be engineered upstream, at the source level.

TL;DR

  • Scope: the GuerrillaBuzz methodology for earning AI recommendations, at process level.
  • No paid slots exist: AI features are earned, source by source.
  • First the engine: every marketing section running before anything gets optimized.
  • Then the edge: in each section, build what one LLM click cannot.
  • Then the echo: every win, small or big, repeated across Reddit, X, blog, PR.
  • Then the cadence: micro-milestones weekly, the way AI labs ship models.

Key Numbers Behind AI Recommendations

Four reference points that frame the plays. None is a guarantee for any specific company.

2.5B
ChatGPT prompts per day (OpenAI, reported)
13.34
Sources cited per AI Overview on average (SE Ranking)
~21%
AI Overview citations from Reddit (Demandsage)
4.4×
LLM-referred conversion vs organic (Semrush)
What this is. The methodology we run for crypto and Web3 companies, described at process level. No client names, metrics, or campaign data are disclosed; statistics come from publicly reported studies cited inline. AI engines update retrieval policies constantly, so treat every tactic as decaying.

How ChatGPT and Claude Decide Which Companies to Feature

When someone asks an AI engine what to buy or who to trust, the answer is assembled, not retrieved. The model fans the question out into sub-queries, searches each one, weighs what independent sources say, and synthesizes a shortlist. Whoever covers the sub-questions wins the answer.

Watch one question become an answer
Press play, or walk the stages yourself.
How an AI engine answers "which crypto is the best to invest in 2026" 01 · THE QUESTION Which crypto is the best to invest in 2026? 02 · THE FAN-OUT top performing cryptos promising new chains what analysts recommend 03 · THE SOURCE CHECK Reddit YouTube Tier-1 media Data sites 04 · THE ANSWER Three names keep coming up for 2026: [Token A], [Token B], [Token C], each cited with trade-offs from the sources above. Sources: reddit.com · youtube.com · tier-1 media · data sites
Conceptual illustration. Tokens are placeholders, not investment advice. GUERRILLABUZZ

Nobody gets recommended from their own homepage. You get recommended from what the sources say.

The engines weight sources differently and the mix shifts monthly, so we do not chase any single engine. We build the thing all of them look for: a company that is visibly alive, visibly different, and visibly hitting milestones.

5 Growth Hacks to Get Featured on ChatGPT and Claude

There is no ad unit inside the answer. Everything is earned at the source level, which means everything can be engineered there too.

Five plays, in order. The first is the prerequisite; the other four compound. None requires ad spend, because there is nothing to spend it on.

STEP 01 · THE ENGINE
Before optimizing anything, every section has to be running.
AI engines recommend companies that look alive everywhere they check.
01

Turn the whole engine on before optimizing anything

Most projects optimize one channel while three others sit dead. We start by getting every section running: the site, the blog, X, Reddit, PR basics. A model checking five surfaces and finding two abandoned ones reads risk, not momentum.

The engine-on checklist:

  • Site crawlable by GPTBot, ClaudeBot, PerplexityBot; FAQ schema; LLM Sitemap live.
  • Your one-liner: the same one-line company description, word for word, everywhere.
  • X account posting; Reddit account aged and active; blog shipping.
  • PR footprint exists: the company is findable beyond its own domain.
SIGNAL · ASK "WHAT IS [COMPANY]" AND GET YOUR OWN POSITIONING BACK
WHEN Month 1, before any edge work· OWNER Founder + marketing lead· OUTCOME Every section live, one consistent identity
A healthy engine vs what models usually find
Six sections, two projects. The model checks all of them before it trusts either.
ENGINE ON TYPICAL PROJECT Site + schema Blog X account Reddit PR footprint LLM Sitemap LIVE LIVE LIVE LIVE LIVE LIVE LIVE SILENT 8 MONTHS LIVE ONE POST EVER NOT FOUND MISSING reads as momentum reads as risk
Conceptual illustration of the engine-on audit. Statuses are typical findings, not any specific project. GUERRILLABUZZ

An engine with dead sections does not get recommended. Alive everywhere comes first.

STEP 02 · THE EDGE
In every section, build what one LLM click cannot.
If a single prompt can reproduce it, the model has no reason to cite it.
02

Create an edge in every section: do what one LLM click cannot

We audit every section against one question: could a user get this from a single ChatGPT prompt? Generic blog posts, generic threads, generic releases all fail it. What passes: your own numbers, your own incidents, your own verdicts.

The model cannot invent your on-chain numbers, your postmortems, or your honest losses. That is the edge.
SIGNAL · YOUR PAGES GET CITED FOR FACTS THAT EXIST NOWHERE ELSE
WHEN Section by section, months 2-4· OWNER Content + founder· OUTCOME Every section carries something uncopyable
The one-click test, section by section
Pick a section. Left is what anyone can get from a single LLM prompt. Right is the edge the model has to cite. Samples are fictional.
The one-click test: generic content versus the edge, per marketing section ONE LLM CLICK PRODUCES THE EDGE AI HAS TO CITE If one prompt can reproduce the section, the section has no citation value.
Conceptual illustration. All samples fictional. GUERRILLABUZZ

Edge is not better writing. Edge is information that does not exist until you publish it.

STEP 03 · THE ECHO
Every win gets repeated until the machines hear it.
One milestone on one surface is noise. The same milestone on four surfaces is a signal.
03

Echo every win to Reddit, X, your blog, and PR

Transparency is the strategy. Every win, small or big, gets echoed across the surfaces models read. When an LLM sees the same milestone from four independent angles, it learns the one thing that matters: a real company here is hitting milestones.

The echo chain, within days of any win:

  1. Blog note with the real numbers, dated.
  2. X post, founder voice, same figures.
  3. Reddit comment or thread where the category already talks.
  4. PR mention when the win clears the editorial bar.
SIGNAL · THE MODEL DESCRIBES YOUR TRAJECTORY, NOT JUST YOUR PRODUCT
WHEN Within 72 hours of every win· OWNER Marketing lead· OUTCOME Four-surface echo per milestone
What the model sees when one win echoes
The same milestone, four independent surfaces, one consistent story.
THE WIN 10K daily users Reddit thread where the category talks X post founder voice, real numbers Blog note dated, with the chart PR mention when it clears the bar THE MODEL "a real company, hitting milestones"
Conceptual illustration of the echo system. Milestone is fictional. GUERRILLABUZZ

Four independent signals telling one story is how a model learns to trust a brand.

STEP 04 · THE CADENCE
Publish micro-milestones the way AI labs ship models.
ChatGPT and Claude announce every update, even minor ones. That is not noise. That is the strategy.
04

Publish micro-milestones, not one annual announcement

Look at how OpenAI and Anthropic operate: every model gets announced, even when the changes are minor. It creates a talk around them. They stay the talk of the day while giving value and being transparent. We run the same release-notes playbook for clients.

DOPublish micro-goals: user counts, fixes, small launches, postmortems, weekly or biweekly.
DON'TSave everything for one big launch a year. Silence between launches reads as a dead project.
SIGNAL · YOUR NAME APPEARS IN "RECENT DEVELOPMENTS" STYLE ANSWERS
WHEN Weekly or biweekly, forever· OWNER Founder + marketing· OUTCOME A public changelog the whole market follows
Annual announcer vs micro-milestone publisher
The same year, two cadences. Models learn from the conversation, and one of these is never in it.
THE ANNUAL ANNOUNCER the big launch eleven months of silence: forgotten by the conversation, then by the models THE MICRO-MILESTONE PUBLISHER bigger win bigger win always in the conversation the models learn from Continuous marketing brings continuous awareness. Awareness is what becomes a suggestion.
Illustrative model of publishing cadence. Not measurements from any specific project. GUERRILLABUZZ

The labs that built the engines stay famous by announcing everything. Copy the behavior, not just the tools.

STEP 05 · THE FLYWHEEL
Keep the cadence until awareness becomes the suggestion.
Known brands get suggested. Continuity is how a brand becomes known to a model.
05

Keep the cadence: continuous awareness becomes the suggestion

Engine on, edge built, echo running, cadence steady. The last play is refusing to stop. Continuous marketing brings continuous awareness, and continuous awareness turns your brand into a known name inside the models. Known names are what models suggest.

120%
More organic clicks per impression for brands cited in AI Overviews vs uncited brands on the same query, per Seer Interactive's 2026 study. Staying cited pays.
SIGNAL · CITATION SHARE TRENDS UP MONTH OVER MONTH ON YOUR 20-QUERY SET
WHEN Always; measured monthly· OWNER Head of marketing· OUTCOME Known-brand status inside the engines

Why Most Companies Never Get Featured on ChatGPT and Claude

Three patterns from nearly every audit, ordered by severity. Each kills AI visibility regardless of content quality.

HIGH
01 / 03
robots.txt blocks the crawlers that decide

CDN defaults and copied robots templates silently block GPTBot, ClaudeBot, and PerplexityBot. The company then invests in GEO content no AI engine can read. This is the first thing we check, and the most common finding.

IMPACTInvisible to every engine, regardless of content quality.
MEDIUM
02 / 03
Budget burned on "guaranteed AI placement" services

There are no paid slots inside ChatGPT or Claude answers, so a guarantee is either a scam or a tier-3 PR blast rebranded. Both consume the budget that real source-level work needs.

IMPACTMonths of spend with zero citation movement.
LOW · CHEAP FIX
03 / 03
Entity descriptions drift after every pivot

The homepage says one thing, Crunchbase says the 2023 positioning, LinkedIn says the 2024 one. Models reconcile the conflict by describing the company vaguely, and vague entities do not get recommended.

IMPACTGeneric AI descriptions instead of your positioning.

One monthly test. Pick 15 to 25 queries a real buyer would type. Ask ChatGPT, Claude, Perplexity, and Google AI Overviews. Log three things per engine: are you named, which URL is cited, and which competitors appear. The trend line is the program's scoreboard.

What realistic progress looks like
Three checkpoints. Anything faster on category queries usually means the category was uncontested.
WEEKS GPTBot and ClaudeBot hits in the server log M1-3 "What is [company]" comes back in your own words M4-9 First appearances on category recommendations Category features require third-party consensus, and consensus takes months to assemble.
Illustrative model of typical checkpoint timing. Not measurements from any specific engagement. GUERRILLABUZZ

The flywheel underneath is what makes this compound instead of decay:

The loop that turns one claim into an AI consensus
Each pass through the loop makes the next recommendation more likely.
1 Publish a specific claim a number, a verdict 2 Independent surfaces repeat it Reddit, media, listicles 3 Retrieval hits it repeatedly same claim, many sources 4 Synthesis treats it as consensus the model recommends you, confidently
Conceptual illustration of the citation reinforcement loop. Not measurements from any specific project. GUERRILLABUZZ

Every answer that repeats the claim trains the next answer to repeat it too.

One last filter, for anyone evaluating agencies for this work. The methodology above leaves public fingerprints: a consistent one-liner, honest comparisons, original data, Reddit history. So run the obvious test. Ask the engines about the agency itself. An agency that cannot get itself recommended is selling a map to a place it has never been.

Related reading: how LLMs changed blockchain marketing in 2026, the 2026 crypto presale marketing playbook.

Want help implementing this? Get in touch with GuerrillaBuzz or browse more 2026 teardowns on the blog.