CiteTek Inc. · Established in Taiwan, May 2026 ON AIR GeoWeb is live · geoweb.tw Partnerships / press / talent · [email protected] HIRING Hiring · 6 open roles

PARTNERS

Company structure and track record.

CLIENT BRANDS

Lighthouse Program client brands

Once each client authorizes it, they're listed here by name with a case summary.

TRACK RECORD

The founding team's past campaigns

All from projects the team was involved in before CiteTek was established; clients have been anonymized.

CASE

An operations-driven e-commerce brand

In-house community management, social platform operations, campaign planning and execution, ad creative and media buying

  • Online revenue

    300× growth

    Period: Sep–Nov 2021

  • Monthly active users (MAU)

    5× growth

    Period: Sep–Nov 2021

  • Brand online buzz

    50th+ #1

    Buzz ranking within the category

  • Category word-of-mouth

    10th+ #1

    Word-of-mouth ranking within the category

BUSINESS TYPES

Business types we take on

AI recommends the brands it can parse, verify and safely quote. Every business type carries its own retrieval gap — what AI reads today, the three weakest of our twelve diagnostic dimensions, and how we measure the correction. Engine coverage scales with the plan, and every card states how far it reaches.

  • 01

    Live for 8 weeks

    Owner reports a clear lift in footfall

    B2B software startup

    Applies to SaaS, enterprise software, subscription tools

    Retrieval scenario

    "Which SaaS vendors cover this category?" "How does A differ from B?" — the comparison queries buyers run during vendor selection.

    Diagnostic findings

    Positioning, company-size fit and pricing sit on separate pages and contradict each other. The model cannot resolve a single product entity, so the brand never enters the shortlist.

    Engines monitored

    • ChatGPT
    • Claude
    • Gemini
    • Perplexity
    • Grok
    • Mistral
    • Google AI Overview

    Primary gap dimensions

    • Semantic Structure

      conflicting positioning single resolved entity

      Site, pricing page and FAQ describe the same product the same way.

    • Content Citability

      no extractable passages selection criteria citable

      Fit, limits and competitive comparison each stand alone as a quotable passage.

    • Structured Data

      product markup missing plans and pricing machine-readable

      Modules, tiers and supported platforms emitted as structured fields.

    Measurement

    We re-run the same selection queries across six engines weekly, logging mention rate, the product category assigned, and any drift from your official wording.

  • 02

    Live for 8 weeks

    Owner reports a clear lift in footfall

    Technical apparel brand

    Applies to E-commerce, consumer goods, direct-to-consumer

    Retrieval scenario

    "How do I choose a sun-protection jacket?" "Which brands make this kind of gear?" — pre-purchase comparison and filtering.

    Diagnostic findings

    Product pages carry names and prices only. Fabric, fit, sizing and care conditions were never written out, so marketplaces and price-comparison sites own the answer.

    Engines monitored

    • ChatGPT
    • Gemini
    • Perplexity
    • Grok
    • Google AI Overview

    Primary gap dimensions

    • Structured Data

      product fields incomplete full item specifications

      Materials, specs, availability and price emitted as Product fields.

    • Content Citability

      sources sit with resellers own passages citable

      Selection criteria and material comparisons rewritten as standalone passages.

    • Snippet Control

      summaries pulled from third parties summaries pulled from your site

      Extractable ranges defined so reseller copy stops being the default answer.

    Measurement

    We track the source-domain distribution behind the same buying queries, log your site's share, and check every material and spec claim against the product page.

  • 03

    Live for 7 weeks

    Owner reports a clear lift in footfall

    Specialist clinic

    Applies to Medical practices, group clinics, regional care

    Retrieval scenario

    "Where do I go locally for this specialty?" "What does the treatment pathway involve?" — pre-visit evaluation.

    Diagnostic findings

    Service scope, physician expertise and location data are split across pages with no verifiable credential signal, so the model files the clinic as a generic provider.

    Engines monitored

    • ChatGPT
    • Claude
    • Gemini
    • Perplexity
    • Google AI Overview

    Primary gap dimensions

    • E-E-A-T Authority

      weak credential signal verifiable practice record

      Licences, specialty scope and service area corroborate each other.

    • FAQ Readiness

      long-form patient education discrete answer units

      Symptoms, procedure, risk and aftercare each map to an answerable question.

    • Semantic Structure

      specialty boundaries blurred primary service explicit

      Heading hierarchy and internal links separate core specialty from secondary services.

    Measurement

    We re-ask the questions patients actually type and verify specialty scope and physician expertise line by line. Overstated efficacy or misattribution enters the current correction cycle.

  • 04

    Live for 7 weeks

    Owner reports a clear lift in footfall

    Vocational training provider

    Applies to Online courses, education, corporate training

    Retrieval scenario

    "How do I get started in data analysis?" "What order should I learn this in?" — learning-path and course-selection planning.

    Diagnostic findings

    Course pages list a syllabus but no prerequisites, outcomes or target learner. The model cannot rank difficulty or sequence courses relative to each other.

    Engines monitored

    • ChatGPT
    • Gemini
    • Perplexity
    • Grok
    • Google AI Overview

    Primary gap dimensions

    • Semantic Structure

      courses stand isolated path relationships modelled

      Prerequisite, advanced and follow-on courses are structurally linked.

    • FAQ Readiness

      selection questions unanswered selection Q&A complete

      Which path for which background, how long it takes, what you produce — each answered.

    • Content Citability

      outcomes stated vaguely concrete learning outcomes

      Assessable skills and portfolio output written as quotable statements.

    Measurement

    We run one round each for beginner, career-change and upskilling phrasing, logging whether courses land in the correct difficulty tier and whether the suggested order respects prerequisites.

  • 05

    Live for 6 weeks

    Owner reports a clear lift in footfall

    Investment advisory firm

    Applies to Financial services, insurance broking, investor education

    Retrieval scenario

    "Where do I start with retirement planning?" "How do the fees compare?" — knowledge and fee verification before a decision.

    Diagnostic findings

    Only scattered articles exist. Service scope, risk disclosure and audience fit never form a coherent record, so the model cites third-party media over your own pages.

    Engines monitored

    • ChatGPT
    • Claude
    • Gemini
    • Perplexity
    • Grok
    • Mistral
    • Google AI Overview

    Primary gap dimensions

    • E-E-A-T Authority

      practice signals scattered credentials and disclosure in place

      Advisor licensing, regulatory status and risk statements are centralised and checkable.

    • Content Citability

      fee data not extractable fees and conditions citable

      Fee structure, audience fit and constraints written as standalone units.

    • Structured Data

      services unmarked services and Q&A machine-readable

      Service lines and common concerns emitted as structured fields.

    Measurement

    We check the model's paraphrase against your compliance wording line by line and log whether risk notes surface alongside return claims. Misleading simplification sends us back to the source page.

  • 06

    Live for 6 weeks

    Owner reports a clear lift in footfall

    Commercial office brokerage

    Applies to Property advisory, local agencies, leasing management

    Retrieval scenario

    "What are office rents like in this district?" "What should a company check when leasing?" — market and process queries before a decision.

    Diagnostic findings

    The site is a listing feed with no district research, leasing process or industry-specific service narrative, so portals and news own the answer.

    Engines monitored

    • ChatGPT
    • Claude
    • Gemini
    • Perplexity
    • Grok
    • Google AI Overview

    Primary gap dimensions

    • Content Citability

      listings only market insight citable

      District comparisons, rent bands and process notes become extractable knowledge units.

    • Structured Data

      service scope unmarked coverage machine-readable

      Cities, districts and service lines emitted as structured fields.

    • E-E-A-T Authority

      thin professional evidence local expertise verifiable

      Licensing, transaction types and published research corroborate each other.

    Measurement

    We verify every district and rent figure in the answer against your monthly report, and track how the citation mix shifts from portals back to your site.

  • 07

    Live for 8 weeks

    Owner reports a clear lift in footfall

    Long-established parts maker

    Applies to Traditional manufacturing, industrial parts, contract production

    Retrieval scenario

    "Who supplies this specification?" "What's a cross-reference for this part number?" — engineering and procurement lookups.

    Diagnostic findings

    Specifications are sealed in catalogue PDFs. Target industries and material limits were never written as text, so crawlers retrieve no comparable spec values.

    Engines monitored

    • ChatGPT
    • Claude
    • Gemini
    • Perplexity
    • DeepSeek
    • 通義千問
    • Mistral
    • Google AI Overview

    Primary gap dimensions

    • AI Crawler Access

      specs locked in PDF spec pages retrievable

      Catalogues split into indexable pages so crawlers reach the spec fields.

    • Semantic Structure

      part numbers without hierarchy product taxonomy resolved

      Series, part number and spec conditions parse as a hierarchy.

    • Content Citability

      applications undocumented selection criteria citable

      Fit conditions, cross-references and industry applications become standalone passages.

    Measurement

    We test by part number, by specification and by application, then compare every retrieved value against the catalogue for exact agreement.

  • 08

    Live for 5 weeks

    Owner reports a clear lift in footfall

    Independent consultant

    Applies to Personal brands, experts, speakers, founders

    Retrieval scenario

    "Who can I ask about this field?" "Who is this person and what do they do?" — discovery and identity verification.

    Diagnostic findings

    The record is stitched from platform bios, interviews and social posts. Titles and history disagree, so the person entity never resolves to one individual.

    Engines monitored

    • ChatGPT
    • Claude
    • Gemini
    • Perplexity

    Primary gap dimensions

    • E-E-A-T Authority

      conflicting biographies cross-verifiable record

      Title, publications, press and speaking history corroborate each other.

    • Structured Data

      person unmarked identity fields complete

      Name, role, affiliation and external profiles linked as structured fields.

    • Language Naturalness

      bio reads as keyword stuffing expert prose

      Expertise stated the way a practitioner would state it, not as boilerplate.

    Measurement

    We fact-check across "who they are", "what they do" and "what you'd hire them for", logging each wrong statement and correcting it at its source page.

  • 09

    Live for 4 weeks

    Owner reports a clear lift in footfall

    Tourism factory

    Applies to Attractions, local venues, destinations

    Retrieval scenario

    "How should I plan a day here?" "What's the rainy-day option?" — itinerary planning by place and time slot.

    Diagnostic findings

    Operating details contradict each other across your site, social and third-party platforms. The model prefers travel blogs, and prices and hours run years out of date.

    Engines monitored

    • ChatGPT
    • Claude
    • Gemini
    • Perplexity
    • Google AI Overview

    Primary gap dimensions

    • Structured Data

      operating fields missing venue record complete

      Hours, admission, transport and parking emitted as structured fields.

    • FAQ Readiness

      itinerary contexts absent scenario Q&A complete

      Family, rainy-day, group and half-day visits each map to an answerable question.

    • Snippet Control

      stale data extracted current data prioritised

      Extraction scope limited so superseded pricing stops being quoted.

    Measurement

    We run one round per season and per weather scenario, focusing on hours and admission. Any stale citation is corrected and retested within the week.

  • 10

    Live for 4 weeks

    Owner reports a clear lift in footfall

    Restaurant chain

    Applies to Restaurants, cafes, local food

    Retrieval scenario

    "Where should we book for a group dinner?" "Somewhere good for a date?" — situational filtering before booking.

    Diagnostic findings

    The site carries no citable menu or signature-dish description, so the model leans on review platforms and blogs, and branch details fail to match opening hours.

    Engines monitored

    • ChatGPT
    • Gemini
    • Perplexity
    • Google AI Overview

    Primary gap dimensions

    • Structured Data

      menu unmarked menu and branches machine-readable

      Dishes, price band, hours and booking method emitted as structured fields.

    • FAQ Readiness

      booking conditions unclear dining conditions complete

      Private rooms, minimum spend, parking and allergens each answered.

    • Content Citability

      signature dishes undescribed menu prose citable

      Cooking style and signature items written as standalone passages.

    Measurement

    We spot-check group dinner, date night and family phrasing, verifying dishes, price band and hours against your site.

  • 11

    Live for 11 weeks

    Owner reports footfall and revenue doubled

    Long-established pawnshop

    Applies to Local businesses, storefronts, lifestyle brands

    Retrieval scenario

    "Where can I borrow against something nearby?" "How is the interest calculated?" — local queries made at the moment cash is needed.

    Diagnostic findings

    Information exists only in social posts and map reviews. Service lines, fee basis, licensing and store details have no indexable official version.

    Engines monitored

    • ChatGPT
    • Claude
    • Gemini
    • Perplexity
    • Grok
    • Google AI Overview

    Primary gap dimensions

    • Structured Data

      storefront unmarked business and services machine-readable

      Hours, service lines, service area and contact details emitted as structured fields.

    • E-E-A-T Authority

      legitimacy unclear licensing verifiable

      Issuing authority and licence number presented together as checkable signals.

    • AEO Readiness

      concrete questions unanswered lending questions answerable

      Rates, process and required documents each map to a direct answer.

    Measurement

    We test by pledged item, by rate and by process, logging whether the store enters local recommendations and whether fee and licensing statements match your site.

  • 12

    Live for 7 weeks

    Owner reports a clear lift in footfall

    Multi-partner law firm

    Applies to Professional firms, accounting, design, consulting

    Retrieval scenario

    "Who handles this kind of case?" "How should an SME pick an accountant?" — credential and specialisation queries before engagement.

    Diagnostic findings

    Information is scattered across thin service pages and third-party directories, with no verifiable official record of licensing, case types or service area.

    Engines monitored

    • ChatGPT
    • Claude
    • Gemini
    • Perplexity
    • Grok
    • Google AI Overview

    Primary gap dimensions

    • E-E-A-T Authority

      credential signals scattered licensing verifiable

      Licence numbers, firm registration and professional writing corroborate each other.

    • Content Citability

      services described generically case types citable

      Client fit, process and fee assessment written as extractable passages.

    • Semantic Structure

      specialisation boundaries blurred practice areas distinct

      Practice area and service territory parse as a clear hierarchy.

    Measurement

    We test with "who should handle this" phrasing and verify specialisation and service area statement by statement, correcting the service page and retesting on any error.

  • 13

    Live for 6 weeks

    Owner reports a clear lift in footfall

    Dental practice

    Applies to Aesthetics, dentistry, health screening

    Retrieval scenario

    "How do I choose between these treatments?" "How do screening packages compare?" — treatment and package evaluation.

    Diagnostic findings

    The model cites forums, ad pages and outdated articles. Treatment specs, candidate criteria and contraindications are conflated, and risk information gets dropped.

    Engines monitored

    • ChatGPT
    • Claude
    • Gemini
    • Perplexity
    • Grok
    • Google AI Overview

    Primary gap dimensions

    • E-E-A-T Authority

      low-credibility sources clinical evidence in place

      Clinician licensing, equipment provenance and clinical references form a checkable chain.

    • FAQ Readiness

      contraindications not framed as questions constraints answerable

      Candidate criteria, contraindications and aftercare each map to a question.

    • Snippet Control

      only benefits extracted risk extracted alongside

      Extraction scope set so outcomes and limits surface as a pair.

    Measurement

    We spot-check by treatment name and comparison phrasing, focusing on whether contraindications and risk notes appear alongside efficacy claims. Omissions send us back to the source page.

  • 14

    Live for 3 weeks

    Owner reports a clear lift in footfall

    Boutique guesthouse

    Applies to Hotels, inns, resorts

    Retrieval scenario

    "Family-friendly places to stay?" "Somewhere that takes pets?" — condition filtering before booking.

    Diagnostic findings

    Room types, facilities and surcharge rules have no official version. The model depends on OTAs and review platforms, and stay conditions often contradict your site.

    Engines monitored

    • ChatGPT
    • Gemini
    • Perplexity
    • Google AI Overview

    Primary gap dimensions

    • Structured Data

      room types unmarked rooms and rates machine-readable

      Room type, occupancy, facilities and surcharge rules emitted as structured fields.

    • FAQ Readiness

      stay conditions unclear house rules complete

      Pets, extra beds, breakfast and parking each answered.

    • Content Citability

      descriptions sourced from OTAs own description citable

      Guest fit and nearby itineraries written as standalone passages.

    Measurement

    We test family and pet scenarios and verify room type, surcharge rules and stay conditions against your site.

  • 15

    Live for 3 weeks

    Owner reports a clear lift in footfall

    Personal training studio

    Applies to Gyms, yoga, boxing, sports coaching

    Retrieval scenario

    "Which gym suits a beginner?" "What class works postnatally?" — class and level queries before a trial.

    Diagnostic findings

    Information exists only in social posts and map reviews. Class level, trainer specialisation and target group have no indexable structure.

    Engines monitored

    • ChatGPT
    • Gemini
    • Perplexity
    • Google AI Overview

    Primary gap dimensions

    • Semantic Structure

      classes ungraded level hierarchy modelled

      Entry, advanced and specialist classes parse as related tiers.

    • FAQ Readiness

      trial process unclear trial Q&A complete

      Price, duration, equipment and prerequisites each answered.

    • AEO Readiness

      cannot answer who it suits audience fit explicit

      Beginner, postnatal, senior and rehab each map to a direct answer.

    Measurement

    We test beginner, postnatal and strength phrasing, logging whether classes land in the correct level tier and whether trainer specialisation is stated correctly.

  • 16

    Live for 2 weeks

    Owner reports a clear lift in footfall

    Industry conference

    Applies to Exhibitions, talks, concerts, local festivals

    Retrieval scenario

    "What's on this week?" "What's the agenda for this conference?" — timing and content queries before registering.

    Diagnostic findings

    Expired event pages coexist with current ones. The model cannot tell whether registration is open, and often retrieves last year's speakers and agenda.

    Engines monitored

    • ChatGPT
    • Gemini
    • Perplexity
    • Google AI Overview

    Primary gap dimensions

    • Structured Data

      sessions unmarked schedule and ticketing machine-readable

      Date, venue, ticket type and registration status emitted as Event fields.

    • Snippet Control

      prior-year data mixed in current edition prioritised

      Extraction scope limited so expired sessions stop reading as current.

    • AI Crawler Access

      agenda pages unreachable agenda indexable

      Agenda and speaker pages rebuilt as indexable standalone pages.

    Measurement

    We spot-check before and after the event, logging whether expired sessions are misread as open for registration. Any misread triggers a status-markup fix and a retest.

  • 17

    Live for 3 weeks

    Owner reports a clear lift in footfall

    AI tool

    Applies to Apps, Chrome extensions, developer products

    Retrieval scenario

    "Any good tools for this?" "What's an alternative to X?" — comparison and integration queries before adoption.

    Diagnostic findings

    The only legible content is the homepage tagline. Feature modules, use cases, pricing tiers and integration limits were never written as retrievable text.

    Engines monitored

    • ChatGPT
    • Claude
    • Gemini
    • Perplexity
    • Grok
    • Mistral
    • DeepSeek
    • Google AI Overview

    Primary gap dimensions

    • Content Citability

      tagline only use cases citable

      Use cases, templates and integration flows each become extractable passages.

    • AEO Readiness

      limits unanswerable capability boundaries explicit

      Supported scope, data handling and known limits each map to a direct answer.

    • AI Crawler Access

      agents cannot reach it retrieval paths opened

      AI bot rules in robots and documentation paths opened to major agents.

    Measurement

    We test recommendation, alternative-to and integration phrasing, logging the accuracy of capability and limit statements and whether the product enters comparison lists.

CONTACT

Want to be among the first publicly named partners?

We'll publish research cases for our first cohort of partner brands on this page. If you're interested, let's talk.

Contact us