03 GroundWork on KOODOS · September 2026

AI Search Visibility

Questions
15
Surfaces
Perplexity · ChatGPT · Gemini · Google AI Mode
Calls
60, no failures
Date
17 September 2026

The baseline

Fifteen questions were put to Perplexity, ChatGPT, Gemini and Google AI Mode. The test used UK targeting where the API allowed it; Gemini could not be UK-targeted. Every call completed. The same question file can now be re-run quarterly without moving the goalposts.

Nobody owns the answer set

webflow.com 6
Semrush 4
HubSpot 4
YouTube 4
Reddit 4

Total citations across the 60-answer baseline for the five most-cited domains or platforms

Everything else appeared only one to three times. Small UK agency blogs were cited beside the tool vendors. The evidence suggests the bar is a specific, UK-framed page that answers the question directly, not blanket domain authority.

What the data says

  1. KOODOS starts from zero

    0 citations or mentions across 60 answers

    The site has no research-led content yet, so the result is unsurprising. Its value is as a repeatable baseline for Coverage.

  2. The citation field is fragmented

    Top domain: 6 citations

    No publisher dominates. Small UK agency articles appear alongside Webflow, Semrush and HubSpot, which makes direct, well-framed answers a realistic route into the set.

  3. Agency recommendations barely overlap

    Different surfaces return different names

    The SEO-audit question produced 22 names across four surfaces; research-led web design produced 15 names with none repeated; AI SEO produced more than 30 names. The AI competitor set is not the same as the Google result set.

  4. A few question spaces are visibly open

    Pre-redesign research and AI recommendations

    What to research before redesigning has no UK small-business answer in the set. How to get recommended by ChatGPT has no settled owner and maps directly to the Coverage offer.

  5. Each surface exposes evidence differently

    Citation counts are not directly equivalent

    Perplexity consistently exposes many sources. ChatGPT sometimes exhausted its response allowance before the citation list. Gemini and Google AI Mode often returned grounded answers with no visible references.

What gets cited by question type

BucketWhat appearedImplication
Problem Small-agency redesign listicles; US SaaS research guides A UK pre-redesign research guide is open
Ownership Webflow, Wix and Squarespace support; platform comparisons Rented Ground and Webflow vs WordPress are the right shapes
GroundWork SEO tool vendors defining an audit as a technical review The market-first definition is absent and belongs on the service page
Coverage Moz and Search Engine Land on GEO; no settled owner for AI recommendations Lead with how businesses get recommended; keep GEO as a spoke
Local Maps listings and directory-style pages Google Business Profile and an Essex page matter more than a local article

The questions behind those gaps

  1. Does my site need a redesign?

    q02 · Problem

    The answer set is dominated by numbered signs-you-need-a-redesign listicles from small agencies. A research-first decision framework can be more useful than another symptoms list.

  2. What should I research before redesigning?

    q04 · Problem

    The cited set is US SaaS guidance from Hotjar, VWO, Figma, Netguru and Impact. No UK small-business page appeared.

  3. What is an SEO audit?

    q07 · GroundWork

    Yoast, Semrush, Sitebulb, Rank Math, WordStream and two agencies frame an audit as a technical review of an existing site. None lead with market demand.

  4. How do I get recommended by ChatGPT?

    q12 · Coverage

    The answer drew Yell, gogomarketing.co.uk, net-craft.com and Entrepreneur, but no owner. It is the exact question Coverage is designed to answer.

  5. Which web design agency should I use in Essex?

    q15 · Local

    Google AI Mode answered with 20 Maps listings. The LLMs also pulled names from directory-shaped sources, reinforcing the case for local profiles and location pages.

Surface behaviour

SurfaceObserved behaviourHow to read it
Perplexity 18-20 sources on every answer The most measurable citation surface
ChatGPT Five answers hit the 1,024-token cap before the citation list Treat those zeroes as truncation, not proof of no sources
Gemini No citations on 7 of 15 answers despite grounding Grounding and visible citations are different measures
Google AI Mode No references on 6 of 15 answers Visible reference counts understate the evidence used

What changes now

  1. Publish the pre-redesign research pillar

    Answers q04 and q02

    Build one GroundWork-voiced guide around what to investigate before a redesign. It should answer the decision, not repeat the existing signs-you-need-a-redesign listicles.

  2. Publish the AI recommendations pillar

    Answers q12

    Explain how assistants choose businesses to recommend, what evidence can influence that choice and how the result will be measured.

  3. Keep GEO as supporting content

    390 searches/mo, but an established publisher set

    Generative engine optimisation remains useful, but Moz, Search Engine Land and Coursera make it the wrong lead pillar.

  4. Use comparison content for ownership

    Webflow vs WordPress

    The cited pattern supports a direct platform comparison as the next Rented Ground spoke.

  5. Put the audit reframe on GroundWork itself

    Market-first, not another generic definition

    The vendor-owned definition space is heavy. The service page should show how KOODOS expands the audit from technical condition to market demand.