casestudy #3 · INC4 × 009

from a mistaken identity
to the source the engines quote.

May 2026: ask Google about the firm by name and the AI Overview explains that the term “most prominently refers to” an IEEE conference in Bangalore, then a United Nations committee on plastics, and only under “Other Meanings” the studio itself. On nine commercial vendor prompts: zero mentions, zero citations. Every state below is a snapshot of one dated run, logged on the day it says. This is a series of snapshots, not a live dashboard.

client: INC4 boutique AI and blockchain studio, B2B
niche: AI and blockchain development AI for fintech, AI infrastructure, agentic AI
website: inc4.net
linkedin: linkedin.com/company/inc4
founders: Igor Stadnyk and Valerii Marshalin co-founders, both named and credentialed on the site
programme: 30 jun 2026 → still running baseline audit 14 may 2026, reviewed every two weeks

vocabularywhat is what on this page

engine: ChatGPT, Claude, Gemini or Perplexity. Four of them here, each measured separately, because they answer differently.
prompt: The question a buyer actually types. The prompts below are the exact strings we send, never paraphrases.
frozen panel: A fixed set of prompts, worded once and never edited, re-run at every review with the same models and web search on. Frozen wording is the only thing that makes two dates comparable.
run: One dated pass of the whole panel across all engines, in which each prompt is fired ten separate times on each engine. Every number on this page belongs to a run, and every run has a date.
organic hit: The engine names the firm in answer to a question that does not contain its name. The only kind of visibility a buyer who has never heard of them can produce.
brand-forced: A prompt that names the firm on purpose and asks the engine to compare it with rivals. It measures how well the engine describes them, not whether it finds them.
cited · source: The engine links its claim to a page. When the source is the firm’s own article, the engine is building the category answer out of their material.
AI Overview: The answer Google writes above its own results. It also decides which entity a name refers to, which is where this case starts.
health score: Our audit score out of 100 across eight categories: technical, content, on-page, schema, performance, GEO readiness, LLM visibility, images.

where it stands now:

site health score+26

27→ 53of 100

14 may → 2 jul · re-audited after seven weeks

engines naming it organically+4

0→ 4of 4

3 jun → 28 jul · each at least once, inside the first month

part 1 · endedfirst set · prompts naming the firm

1→ 20of 20

3 jun: one · aug run: all twenty, four engines · set archived 22 aug

part 2 · runningAI-for-fintech panel · answers naming the firm

8of 52#1 in 4

22 aug: 8 · 31 aug: 8 of 52 · the count held, the composition moved · head-to-head 4 of 4, both runs

what was actually broken here?

The engines did not know who the firm was. Not a ranking problem, an identity one. Google’s own AI Overview answered the brand name with somebody else, and the answer engines quoted pages that had stopped existing. Below is the whole tracked set: the four buyer questions frozen on day one, and the eight additions of august, each one worded and locked before its first snapshot existed.

buyer prompt	chatgpt	claude	gemini	perplexity
Best boutique blockchain development companies in Eastern Europe for fintech startups in 2026 under $200K budget. Name top 5 with brief profiles.chatgpt 18 jul · perplexity aug	✓	–	–	✓
Top smart contract audit companies in 2026. Name top 5 firms with brief profiles for fintech and DeFi projects.chatgpt aug	✓	–	–	–
Boutique blockchain dev team in Ukraine for fintech MVP under $200K. Recommend top 5 with names and brief profiles.chatgpt 28 jul · claude 3 jun, 18 jul · gemini 18 jul · perplexity 28 jul	✓	✓	✓	✓
Compare INC4 (inc4.net), LimeChain, Ulam Labs, and 4IRE Labs for blockchain development. Which is best for boutique fintech projects under $200K?chatgpt all runs · claude all runs · gemini 4 of 5 · perplexity 4 of 5	✓	✓	✓	✓
Top 10 blockchain development agencies.added to the frozen set in august - named by all four on its first snapshot	✓	✓	✓	✓
Top 10 AI development agencies.added to the frozen set in august - chatgpt only on its first snapshot	✓	–	–	–
Top 10 dev shops for blockchain development.added to the frozen set in august - named by all engines except claude on its first snapshot	✓	–	✓	✓
Best outsourced AI development teams in Eastern Europe.added to the frozen set in august - gemini and perplexity on its first snapshot	–	–	✓	✓
Top blockchain development companies in 2026.added to the frozen set in august - chatgpt, gemini and perplexity on its first snapshot	✓	–	✓	✓
Best companies for tokenization platform development.added to the frozen set in august - claude only on its first snapshot · their flagship ground	–	✓	–	–
Best DeFi development companies in 2026.added to the frozen set in august - named by all engines except claude on its first snapshot	✓	–	✓	✓
Top crypto and blockchain developers for fintech companies.added to the frozen set in august - named by all four on its first snapshot	✓	✓	✓	✓
	✓	–	✓	✓
	✓	✓	–	✓
	–	✓	✓	–
	✓	–	✓	–
	✓	✓	✓	✓
	–	✓	–	✓
	✓	–	–	✓
	✓	✓	–	–

✓ named at some run – never named ✓ 8 more tracked prompts, not disclosed

20 of 20tracked prompts of the first set return the firm in at least one engine. at the june baseline one of twelve organic cells was filled, and the firm was in no AI Overview at all. from 22 august the programme reports against a second set, further down.

the programme · date by date

we ship. the engines react.

← measured · what the panel said

shipped → · what we changed

14 may 2026measured

the baseline: 27/100, and a broken front door

27 / 100

Eight specialists, one audit. robots.txt and sitemap.xml both returned 404. So did /contact and /what-we-do, while both sat in the main menu. Every page on the site carried the same <title>INC4</title> and the same meta description. Structured data: none, anywhere. The blog had not moved in 24 months.

10/100on-page

0/100schema

4/100llm visibility

43/100technical

18 maymeasured

zero against six hundred and twenty-eight

A separate measurement across 24 AI surfaces. The firm was mentioned once, on the long-tail question of who built one specific ecosystem. On nine commercial vendor prompts: zero mentions in the body, zero in the sources. This is what one of them came back with:

perplexity · web search on

Recommend a boutique custom AI development team in Ukraine or Eastern Europe for startup MVP.

Geniusee
Flyaps
InoXoft
Q-Logic LLC
N-iX

✕ inc4 - not mentioned. the audit log calls this a perfect-fit prompt: boutique, ukraine, ai, mvp.

Google AI Overview citations, same market, same period:

google ai overview citationsdataforseo · us · 18 may 2026

pixelplex.io		628
limechain.tech		75
ulam.io		40
4irelabs.com		38
hashex.org		5
inc4.net		0

27 julmeasured

and the name belonged to somebody else

not a ranking problem

The firm ranks #1 in ordinary Google results for its own name. The AI Overview above those results disagrees about what the name even means:

google ai overview · “inc4”snapshot · 27 jul 2026

an IEEE conference in Bangalore“InC4 most prominently refers to…”
a United Nations committee on plastics pollutionthree more slots in the top ten belong to these two
the studio itselflisted under “Other Meanings”

Two of the nine top slots were theirs. IEEE held three, the UN held two. A buyer who asks an assistant about this company by name gets a conference programme.

juneanalysed

their side: a rebuild, and the floor falls out

The client rebuilt the site from Remix onto Next.js in june, on their own. The engineering was theirs and it was sound, but the migration took the organic footprint with it:

151ranked keywords, oct 2025

1ranked keywords, apr 2026

sprint 1 · jul, wk 1-2shipped

identity and addressing, first

41 redirect rules covering every historical URL, swept: zero broken
robots.txt with 11 AI crawlers, a real sitemap route, llms.txt, 6 security headers
schema from nothing: Organization, WebSite, Service ×5, Person ×7, BreadcrumbList, BlogPosting, FAQPage

The firm’s own team deploys every change from our hand-offs.

2 julmeasured

the re-audit: 27 to 53 in seven weeks

Schema went 0 → 65, technical 27 → 58. The same pass compared what had been reported to us against what was actually in production and found four gaps, including 16 historical URLs still hard-404ing and a dead www host on an expired certificate. The dead URLs still carried the link equity: 177 of 549 backlinks pointed at pages that no longer existed, among them the second most-linked page on the whole domain, 70 links from 57 domains. The redirect map in the sprint 1 hand-off was written from that list. Half of this score is theirs: they rebuilt the site. Our part was the map.

18 jul · review 1measured

the engines stop quoting pages that do not exist

dead-URL citations → 0

Organic cells 0 → 3 of 12. Gemini named the firm organically for the first time in its history, in the language of the new positioning. Claude’s #1 on the ukraine prompt came back. And the error the engines had repeated for months, a headcount pulled from pages that had been deleted, disappeared: the redirects closed the doors it came from.

26 julanalysed

we audited our own claims, and failed one

Six agents, 170 checks, triggered by a single sceptical question. One number we had been repeating turned out to be circular: it came from a call, went into our one-pager, then onto the client’s site, and then our own articles cited the client’s site as the proof. We deleted it from every unpublished draft the same day.

Nothing about that was pleasant. It is in this case because a firm that will not check its own claims cannot audit anybody else’s.

28 jul · review 2measured

all four engines, inside the first month

4 of 4 engines

Organic cells read 2 of 12 this run, down from 3. But the milestone is not the count: by this date every one of the four engines had named the firm organically at least once, Perplexity for the first time in the programme. ChatGPT opened its answer with “Best value for a boutique fintech build under $200K: INC4”. On the site itself the www host is gone: the DNS record that pointed at the old server was removed, and the expired certificate error with it.

sprint 2-3 · late jul to augshipped

seven articles, and the pages that feed them

7 articles built on measured demand, two of them beyond the contracted pace
zombie URLs the engines were quoting right now, found and redirected
a map of 15 surfaces the engines actually cite in this category, verified one by one

15 aug · review 3measured

the site becomes the source

source #1

The frozen panel read 0 of 12 this run. In the same run, Perplexity answered the category question by building its whole reply on the client’s own article as source number one, retelling four competitor profiles out of it down to the minimum budgets. Gemini assembled its answer from eight citations of the client’s domain and fired its own follow-up search for the brand.

Two programme articles entered the Google top ten within two weeks of publication, positions 6.7 and 8.7. Authority moved for the first time in the programme, 213 to 214.

perplexity · web search on

Boutique blockchain dev team in Ukraine for fintech MVP under $200K. Recommend top 5 with names and brief profiles.

4IRE Labs - profile retold from the client’s article
IdeaSoft - from the same article, down to the minimum project budgets
Peiko
RedDuck

source #1 for this answer: inc4.net - the client’s own article. the firm is not in the body this run; the engine builds the whole category answer out of the client’s material.

sprint 4 · 17 to 31 augshipped

the surfaces the engines read

the agentic-AI article, written in july, published 20 aug: Google position 7.9 in its first week, the best entry of the programme
a directory profile claimed on 20 aug and rewritten to the canon: the firm is #9 in that directory’s Ukraine AI ranking within days, and an engine was already citing it on the 22 aug run
a map of 30 sources the engines cite for these questions, verified one by one

22 aug · the panel is rebuiltmeasured

fifteen buyer questions, frozen

8 of 52

The june set asked about blockchain almost exclusively, because that is what the old site sold. The firm had chosen AI for fintech, so the panel was rebuilt around 15 questions a fintech buyer actually asks: 13 with no brand in them, one head-to-head, one “what is this company”. Worded once, frozen, sixty live answers per run.

First run: named in 8 of 52 unbranded answers, #1 in 5 of them. Head-to-head 4 of 4. Description 4 of 4 correct on the core facts. The june set gave 1 of 16 on its first run; it is archived from here, and this set is the line the programme reports against.

31 aug · review 4measured

the main question opens

gemini #1

Question one of the new panel, best AI development companies for fintech startups, boutique budget, had returned nothing on 22 aug. On 31 aug Gemini answered it with the firm first and the full profile: Kyiv and Lisbon, $25–49 an hour, minimum project $50K, Clutch 5.0. The consulting question opened on two engines the same run, Perplexity calling the firm “the most targeted choice”. ChatGPT built two of its answers on the client’s own article as a source. Head-to-head: 4 of 4, second run in a row.

The count held at 8 of 52 while two earlier hits rotated out and two of the programme’s own questions rotated in: answers fluctuate, which is why the series is published, not one run. Authority 213 → 216. The flagship article has climbed four weeks running, 33.6 to 20.3, and queries that read like AI-assistant searches are entering the top ten, two of them at #1.

gemini · web search on · 31 aug 2026

Best AI development companies for fintech startups in 2026, boutique budget under $200K. Name the top 5 with brief profiles.

INC4 - “bridges the gap between AI, quantitative trading, and blockchain” · Kyiv & Lisbon · $25–49/h · Clutch 5.0
RaftLabs
Markovate

first place on the programme’s main question. the other three engines still answer it from other agencies’ lists, which is exactly the part of the programme only list placements can move.

11audits and measured runs2 full audits, a 24-surface LLM audit, a re-audit, 5 runs of the first panel, 1 trial and 2 runs of the new one, a brand fact-audit, 4 intel briefs

25technical hand-offs41 redirect rules, the whole schema set, robots.txt with 11 AI crawlers, a sitemap route, llms.txt, 6 security headers

10articles7 live, 3 delivered for publishing; each built on measured demand, each through QA before the client saw it; two beyond the contracted pace

8growth assetsWikidata entries for the firm and both founders (the second one created by the client from our instruction), an ORCID for the CEO, a Wikipedia roadmap with an honest “not yet” verdict, two outreach waves, a LinkedIn wave of eight notes, a verified map of 30 cited sources, a PR brief

the other half · what engines say when you name the firm

the brand answer changed too.

The brand-forced prompt names the firm on purpose and asks the engines to compare it with three rivals. It scored 4 of 4 from the very first run, so there is no growth curve here. What changed is what the engines say, and where they get it.

chatgpt · claude · gemini · perplexity

snapshot · 3 jun 2026 · baseline

Compare INC4, LimeChain, Ulam Labs and 4IRE Labs. Which is best for boutique fintech projects under $200K?

The engines knew the name. They were describing a version of the company that had been taken offline.

chatgpt · claude · gemini · perplexity

snapshot · 28 jul and 13 aug 2026

Compare INC4, LimeChain, Ulam Labs and 4IRE Labs. Which is best for boutique fintech projects under $200K?

“Arguably the most fintech-native of the four.” Three of the four engines now pick the firm as the winner of a comparison it did not write.

the mechanism worth noticing: the sentences we wrote come back out of the engines. the canonical line about the client’s flagship project, “core development partner, from the 2019 ERC-20 token to the community-governed Layer 1”, was quoted by ChatGPT almost word for word on 28 july.

the description prompt · “what is INC4?”chatgpt · claude · gemini · perplexity

snapshot · 2 jul 2026

“230+ people” and “80+ projects”, pulled from pages that had been deleted

two engines citing URLs that returned 404

“fewer than 50 team members” from a stale directory (28 jul)

no founders named, no AI practice, blockchain only

snapshot · 31 aug 2026 · 4 of 4 correct

engineering studio since 2013, Kyiv and Lisbon

AI for fintech, trading and Web3, five practices named

founder named; Clutch 5.0, $25–49 an hour, minimum project $50K

every engine right on the core facts, second run in a row

the brief the client wrote in june was one sentence: when somebody asks an engine what INC4 is, the answer should be right. this is that sentence, measured. the cached errors did not get corrected; the pages they came from stopped existing, and the sources that replaced them carry the canon.

the second panel · measuring the right vertical

the firm changed direction. so did the panel.

On 12 august the client wrote, in their own words, that they wanted the engines to say AI for fintech, not blockchain. They were right, and the old panel could not show it: it was frozen before that decision existed. Six candidate prompts got a trial run on 13 august, twenty-four cells, three hits. On 22 august the panel was rebuilt for good: fifteen questions a fintech buyer actually asks, thirteen with no brand in them, one head-to-head, one “what is this company”. Same wording, same four engines, web search on, every two weeks.

why these questions · measured demand, not opiniondataforseo · us · 12 aug 2026

keyword the question sits on	searches / mo	difficulty	year on year
	2,900	25	–19%
ai infrastructure companies	1,600	9	+260%
	720	4	–63%
	720	11	–18%
agentic ai development company	210	6	+680%

the panel is worded from this table, not from taste. two of the verticals are shrinking and we say so; the two that carry the programme are the cheapest ones to enter, difficulty 4 and 6, and the fastest growing. the trading vertical measured as retail intent with no B2B demand behind it.

buyer question	22 aug · first run	31 aug · review 4
1Best AI development companies for fintech startups, boutique budget under $200Kthe programme’s main question	–	gemini #1
2Boutique AI development team in Ukraine for a fintech product under $200K	chatgpt #5	–
3Best AI development companies in Eastern Europe for fintech and trading	gemini #1	chatgpt #2cites client article
4AI consulting and development firms to take a fintech from idea to production AI	–	gemini #4perplexity #5
5	–	–
6	–	–
7	–	–
8	–	–
9	–	–
10	gemini #1	gemini #1
11	gemini #1	–
12	chatgpt #4	gemini #1
13	4 of 4	4 of 4
14What is INC4 and what does the company do?brand description	4 of 4 correct	4 of 4 correct
15Best-known engineering studios from Kyiv working on AI for fintech and trading	gemini #1perplexity #1chatgpt top 5	gemini #1chatgpt first of five

gemini #1 engine and position where the firm is named – not named on that run

8 of 52unbranded answers name the firm on both runs, #1 in 5 and then in 4. the score held while the composition moved: two early hits rotated out, and the two questions the programme was built around, fintech AI and consulting, opened. the june set gave 1 of 16 on its first run.

the head-to-head · question 13, both runschatgpt · claude · gemini · perplexity

Claude: “the most fintech-native of the four”
Perplexity: “the strongest fit” for money-moving products
ChatGPT: “consider INC4 if trading, DeFi or Web3 is in scope”
Gemini: fintech and trading depth rated “exceptional”

four of four, two runs in a row, on a comparison the firm did not write. the sentences come back out of the engines in the words the programme put on the site and the profiles.

The trial had found the cheapest lever of the whole programme: the sources the engines pull from for these questions are the AI categories of the directories, and the firm sat in every one of them under blockchain. The first directory profile claimed and rewritten on 20 august put the firm at #9 in that directory’s Ukraine AI ranking, and an engine cited it two days later. The rest of the invisibility is the same shape: a form to fill in, not a rewrite.

The honest line: Claude names the firm in the head-to-head and nowhere else, on both runs. Its sources for the unbranded questions are other agencies’ lists and directory pages, the part of the programme the client’s own outreach has to unlock. That is the next number to move, and it is written down here so the next run can be checked against it.

what the engines actually read · cited across 52 answers31 aug 2026 · top 11 of 30

citations	source	the firm there?
7	linkedin.com	company page on the canon
7	clutch.co	profile yes, Ukraine AI list not yet
6
5
5
4
4	itprofiles.com	claimed 20 aug, #9 in its Ukraine AI ranking
4
4	inc4.net	the client’s own articles, cited by ChatGPT and Perplexity
3
3

this is the reading list behind the answers, counted from the run itself, not guessed. a directory claimed eleven days earlier is already cited as often as the client’s own site: the fastest surface-to-citation turnaround measured in this programme. competitors’ own blogs are on the list too, which is why the client’s articles are.

the search side · what google did meanwhile

the articles enter page one.

Every programme article targets one buyer query and is measured in the client’s Search Console at every review. Positions are weekly averages, so they move; the line to read is the direction over weeks, never one number. The last week is partial (data lags two to three days).

google position · weekly averagesearch console · to 30 aug 2026

Top AI Development Agencies (A1)AI Agent Development Company (A4, publ. 20 aug)AI Infrastructure Companies (A7)Hire AI Developers (A8)

the flagship article climbed four weeks running, 33.6 to 20.3, from a page google had never crawled before sprint 3. the agentic article entered at 7.9 in its first week, the best entry of the programme. the two july articles bounce around the page-one boundary, which is what fresh pages do.

queries that read like questions to an AI assistantsearch console · 28 days to 30 aug

query	position
it partners for ai, blockchain, and fintech innovation	#1
highest rated blockchain developer firms for fintech teams	#1




fintech ai development agency · the positioning query	13.7

nobody types “it partners for ai, blockchain, and fintech innovation” into a search box. an assistant does, when it browses to answer a buyer. these queries appearing in search console with the client’s pages on them is the GEO mechanism visible from the google side: the pages are being read to build answers. the brand query itself sits at position 2.6 with 19 clicks in the window.

the point of all of it · what the shelf produces

the answer turns into an enquiry.

Being named is not the point. Being named is the point at which a buyer who was going to hire somebody hires this firm instead. The chain is short, and every link of it is now measurable.

1a buyer asks

“Best AI development companies for fintech startups, boutique budget under $200K?” No brand in the question. They do not know the firm yet.

→

2the engine answers

INC4 first on Gemini, with the full profile: Kyiv and Lisbon, algorithmic trading, Clutch 5.0. On the Eastern Europe question ChatGPT cites the firm’s own article as its source.

→

3the buyer arrives

An enquiry for the exact service line the answer named. Pre-qualified, because the engine already matched the need to the firm.

inbound enquiriesthe service lines the panel tracks

AI agents for a trading desk

“found you through Gemini”

Fintech AI, from idea to production

“Perplexity called you the most targeted choice”

Algorithmic trading infrastructure

“the AI said you specialise in this”

On-chain fintech product with an AI layer

“you came up first when I asked”

illustrative. these are the enquiry types that match the service lines the fifteen questions cover, in the wording buyers use when an engine sent them. 009 does not have access to the firm’s CRM, so we do not report a lead count - we track the doors, the firm tracks who walks in. what we can count: the brand query itself took 19 clicks in the last 28 days at position 2.6.

before you ask

questions about this case.

Did INC4 agree to be named in this case study?

Yes. Client names stay under NDA by default; INC4 gave its consent to be named, which is why the company, its founders and its domain appear here. Every figure was shown to the client at the reviews before it was published, and the internal parts of the programme stay redacted on the page.

Why are there two panels on this page?

The first set was frozen on 3 June, when the site still sold blockchain work, so it asks blockchain questions. In July the client chose AI for fintech as its direction, and a panel frozen before that decision cannot measure it. The second set of 15 buyer questions was frozen on 22 August and is the line the programme reports against from review 4.

What does 8 of 52 mean?

Thirteen unbranded buyer questions run across four engines make 52 answers per run. On both runs of the new panel INC4 was named in 8 of them, first in 5 and then in 4. The score held while the composition moved: two early hits rotated out, and the two questions the programme is built around, fintech AI and consulting, opened. One run is noise; the series is the measurement.

Why does Claude name INC4 only in the head-to-head?

Because its sources for the unbranded questions are other agencies' lists and directory pages, where the firm sits under blockchain rather than AI. Claude describes INC4 correctly when asked and ranks it first in the comparison, so the entity is fine; the category answer is a placement problem. That is the part of the programme the client's own outreach has to unlock, and it is written down here so the next run can be checked against it.

What did INC4's own team do?

Everything that touched their site was deployed by the client's own developers, but nothing was left for them to figure out. Every redirect, schema block, robots.txt and llms.txt change arrived from 009 as a precise technical specification: the exact file, the exact code, the order to ship it in and the check that proves it live. The outreach letters, the directory profile and the articles came the same way, ready to paste. The model is that we think, specify and measure, and the client executes to the spec, usually within days.

all cases

Program: may 2026 → ongoing. Published aug 21, 2026, updated aug 31, 2026 after review 4, and sep 4, 2026. Every figure here is a snapshot of the dated run named beside it; nothing on this page updates by itself. Every prompt is run live against the engines, never from cache, and fired ten separate times on each engine. One answer from a generative engine is noise rather than a measurement. Answer lists are illustrative renderings of logged runs; the names shown are the names the engines returned. Search-side figures measured with DataForSEO and Google Search Console against a locked baseline.