METHODOLOGY

Last updated 11 November 2026.

How sAIph Measures AI Search Visibility

Real guest questions in five live engines. Every answer read by a person. A published score formula, and anything we could not measure labelled as such.

AT A GLANCE

sAIph Search measures AI visibility by testing real guest questions in ChatGPT, Gemini, Claude, Perplexity, and Copilot.

A person reads every answer and labels it Featured, Mentioned or Not present. The result is an AI Visibility Score from 0 to 100, and every report states how many questions, engines and runs it rests on.

5live AI engines, every answer read by a person

0 to 100AI Visibility Score, re-measured every month

Every answerread by a person

Trusted by world-renowned five-star hotels including The Langham, London and Raffles London at The OWO.

See where your hotel stands, free →"There's nothing they need to improve."Joe Bamsey, then Director of E-Commerce, The Langham, LondonRead Joe's review on Clutch ↗

THE SCORE

One formula, published.

AI Visibility Score = (Featured × 1.0 + Mentioned × 0.5) ÷ total questions × 100

Featured

You are the main or explicitly recommended answer.

Mentioned

You appear, but are not the recommendation.

Not present

You are not named at all.

See this method run on your own hotel, free.

Ten real guest questions, your competitors side by side, within two working days.

Get my free snapshot Ask us a question

HOW WE CHOOSE QUESTIONS

From your revenue, not a keyword list.

✓Three to six revenue priorities agreed with you, weighted by what earns

✓Short questions, phrased the way guests type them

✓Most never name you, because those test whether guests find you

✓Each could fairly be won by at least five real competitors

✓You approve the list before testing starts

RUNS

	Snapshot	Audit	Programme
Questions	10	100+	Your set
Engines	Live engines	Five	Five
Re-test	Not applicable	Day 30	Monthly

WHAT WE DO NOT CLAIM

What we do not measure

No single-run rates

AI answers change between runs. Single runs are labelled snapshots, never rates.

Modelled AI Overviews

Google AI Overviews are modelled from Search Console data and labelled as such.

No internal rankings

We do not claim to see inside the engines' ranking systems.

Engine changes noted

A major model release triggers a re-check, logged in every report.

WHAT'S CHANGED IN AI SEARCH

What has changed in AI search this month?

The engines change their sources and rules every few weeks. We track what moved, what we have stopped recommending and why.

Last reviewed 1 October 2026. Updated monthly.

Every change, with sources →

What we have stopped recommending, and why

Tactic	What changed	What we do instead
llms.txt as a visibility lever	97% of files receive no requests (Ahrefs, Jun 2026)	We no longer include it; the time goes on content, access and sources
Schema as a citation tactic	No measurable citation lift in controlled testing	Schema for clarity; complete, visible facts on the page
FAQ markup for search snippets	Google withdrew FAQ rich results in May 2026	Visible answers to real guest questions
Reddit seeding for ChatGPT	ChatGPT's Reddit citations fell sharply in August	Earned coverage and accurate listings; genuine, disclosed answers on Reddit kept for Perplexity
Content chunking or rewriting pages "for AI"	No evidence it helps; risks thin pages	Pages written for guests, with the facts stated plainly

AI search changes faster than most review cycles. We re-test our own advice every month.

THE SCORE

How is the score weighted?

Full audits score eight modules, weighted by how directly each decides whether AI names you.

Module	Weight
AI visibility (share of voice)	28
Entity, authority and off-site presence	19
Content substance and extractability	12
Technical crawlability and rendering	11
Machine-legible entity and attribute data	10
Answer quality and stability	9
Fetchability and response reliability	7
Advanced (multilingual, agentic and measurement readiness)	4

Critical findings come first and cap the score: AI crawlers blocked at the firewall, or key pages that only render with code, cap it at 40. False facts about you, or no visibility on questions that do not name you, cap it at 60. We do not give letter grades, because a grade implies a comparison with other hotels that no one has published.

Not sure where to start? Ask us anything.

Advice, a second opinion or a quick question. We reply within one working day.

Get in touch

REPORTING

What do you report?

Visibility. How often each engine names you on questions that do not mention you.
Share of voice. Against the competitors you name and the ones the engines bring up.
Accuracy and sentiment. What the engines say about you, and whether it is true.
Sources. Which pages and third-party sites the engines rely on.
Business outcomes. From GA4 and Search Console where you share them.

A single run is a snapshot, not a rate. In SparkToro's testing, fewer than 1% of repeated runs returned the same list of brands, so we use repeated runs wherever a decision depends on the number.

Source: SparkToro, AI recommendation consistency research, Jan 2026.

Planning has moved before booking has: most guests still book on a website, which is why being the hotel AI describes first turns AI discovery into a direct booking. Only 8% of UK and US travellers currently rely on AI chatbots to plan, and 66% would not let an AI book for them; under 1% of Booking Holdings room nights come from conversational agents.

Source: Expedia Group with YouGov, 5,700 adults, Mar 2026; Booking Holdings, Aug 2026.

ENGINE CHANGES

How do you handle engine changes?

When an engine changes materially, for example with a new model release, we re-check your questions straight away, log the change on What's changed in AI search and note the break in your report. Re-tests keep the same questions, engines, run count and scoring version. If a change lands inside a guarantee window, we pause and re-baseline.

The key takeaway: every sAIph figure tells you what it rests on, and anything we could not measure is labelled as such.

FREE AI SEARCH VISIBILITY SNAPSHOT

See who AI recommends instead of you, free.

You against your competitors · What AI says about you · Up to five quick wins your team can make this month · Where rivals are winning. Free, within two working days.

Two emails: your Snapshot and one check-in a week later. Nothing else unless you reply. We never sell your details. Privacy notice →