Checking the evidence

How to measure AI visibility without a made-up rank

By Ranklore · First published 6 September 2026
This page is maintained periodically. Last updated 26 October 2026.

Why rank position alone is no longer enough

In February 2026, Ahrefs re-ran their 300,000-keyword click-through rate study using December 2025 Google Search Console data. When an AI Overview appears on a query, the top organic result's click-through rate drops by 58%. That is nearly double the 34.5% reduction Ahrefs measured eight months earlier in their April 2025 study using the same methodology. Position 2 loses 50.8% and position 3 loses 46.4%. Ahrefs CTR study, February 2026, summarised by Auspia and Am I Cited; methodology detail.

The 58% figure is not a single-source claim. Five independent studies using different methods all find a significant reduction:

Sources: Axis Intelligence compilation; PPC Land coverage of ISB/CMU paper; Analyze.ai cross-study comparison.

The zero-click picture is starker. SparkToro's analysis of Similarweb clickstream data covering January to April 2026 found that 68.01% of US Google searches end without a single click to any website, up from 60.45% in 2024. SparkToro zero-click study, 2026. When an AI Overview is present, the zero-click rate jumps to 83%. In Google's full AI Mode, it reaches 93%. Axis Intelligence AI search statistics. Gartner predicts organic search traffic to websites will fall by 25% by the end of 2026. Gartner prediction, cited in industry compilations.

What this means for measurement. If ranking position 1 loses 58% of its clicks when an AI Overview appears, then tracking "we moved from position 5 to position 2" no longer tells you whether traffic or enquiries increased. A page can rank higher and receive fewer clicks because the AI Overview absorbed the answer. Conversely, a page that ranks position 7 but is cited inside the AI Overview can receive more qualified traffic than a position 1 page that is not cited. Traditional rank tracking measures a signal that is increasingly decoupled from business outcomes.

This is why AI visibility measurement (mention rate, citation rate, sentiment, position within the answer) is not an optional extra in 2026. It is the layer that explains why rankings changed without traffic following.

Recovery and uncertainty. Organic CTR on AI-Overview queries partially recovered from a low of 1.3% in December 2025 to 2.4% by February 2026, an 85% increase, but this remains 37% below the pre-AI-Overview baseline of 3.2%. theStacc recovery analysis. All studies cited use US or European data; no Singapore-specific CTR impact study exists as of September 2026. The French market showed a smaller 23.1% reduction, suggesting market-level variation. Ahrefs France impact study, September 2026. Singapore SMEs should treat these figures as directional evidence of a structural shift, not as precise local predictions.

Does a mention count as a citation?

A business name appearing in an answer is a mention. A source link is a citation to a page. OpenAI explains how to open citations and source links in ChatGPT Search answers. Follow the link and check whether the destination supports the statement.

For the register below, count a link to your own website as an owned-site citation. Record a directory, publisher or other third-party page separately. These are reporting definitions: a citation is not proof that its statement is correct or that it generated an enquiry.

Why repeat the check?

In SparkToro's January 2026 study, repeated brand-recommendation prompts produced varying lists and orderings. The researchers found appearance frequency more informative than treating answer order as a stable rank. The study's prompt selection and provider context limit how broadly its findings apply. Read the study and its methodology.

For your own checks, save the exact question, engine, date, market context and response. Repeat the same version before comparing periods. Changing the question while drawing a continuous trend hides a method change.

What should the sample register contain?

AI observation register:

A worked example

Illustrative numbers: suppose 30 valid answers include six mentions of a business, and three of those answers link to its website. The mention rate is 6/30, or 20%. The owned-site citation rate is 3/30, or 10%.

Those figures describe the valid answers in that test. They do not mean 20% of customers saw the brand, or that 10% of enquiries came from AI. Report failed attempts separately and explain what counted as a valid response. Small, correlated prompt samples cannot support confident market-wide claims.

What should an AI visibility audit cost in Singapore?

The "AI visibility audit" search results are dominated by SaaS tools and PDF reports, not transparent agency pricing. No Singapore agency publishes a clear, itemised price for a standalone AI visibility audit as of October 2026. This is a pricing gap.

October update: two Singapore SEO practitioners now publish unusually transparent SEO retainer pricing, but neither prices a standalone AI visibility audit. Nick Tung's June 2026 breakdown puts Singapore SEO at S$500 to S$20,000 per month across five tiers and states that hidden costs (content, technical remediation, tools) typically add 1.2 to 1.5 times the headline retainer. Nick Tung: SEO agency pricing Singapore 2026. Ren Hao SEO's June 2026 guide puts managed SEO at S$1,500 to S$5,000 per month, with the median SMB retainer around S$2,500 to S$3,500, usually quoted plus GST. Ren Hao SEO: what drives SEO costs in Singapore. Both guides are about retainer SEO, not one-time AI visibility audits, which is why the standalone audit pricing gap remains.

A 23 October 2026 update adds one more data point. MissLobang's agency roundup names Awebstar as the Singapore agency publishing the cheapest itemised SEO tiers: S$800/month for up to 20 keyphrases, S$1,400 for up to 60 and S$2,300 for up to 120, with AI citation tracking across ChatGPT, AI Overviews and Gemini included only on the top tier. MissLobang agency roundup, 23 October 2026. That is visibility monitoring bundled into a premium retainer, not a standalone audit price. The gap stands.

International benchmarks provide a range. Surferstack's September 2026 analysis of seven productised GEO service packages found that one-time AI visibility audits anchor between US$750 and US$2,500, typically including baseline visibility scores across ChatGPT, Perplexity and Google AI Overviews, a competitor comparison and a prioritised fix list. Surferstack: GEO service packages, September 2026. Soar.sh's May 2026 pricing guide places one-time GEO audits between US$1,500 and US$7,500, with monthly retainers from US$1,500 to US$25,000+. Soar: AI visibility agency pricing 2026. The Remarkable Agency's August 2026 guide cites one-time audits at US$1,500 to US$5,000. The Remarkable: AEO/GEO agency cost, August 2026.

In Singapore, the picture is less transparent. MediaOne sells GEO as a monthly retainer starting at S$580 for one AI platform, with no standalone audit price listed. MediaOne GEO pricing, updated October 2026. AI Studio Singapore's May 2026 agency roundup estimates budget GEO services at S$550 to S$800 per month, mid-range at S$1,500 to S$2,500, and enterprise above S$5,000, again as retainers rather than one-time audits. AI Studio Singapore: best GEO agencies, May 2026. Wikibusines notes that US$500 once buys a full AI-visibility audit, while US$500 per month buys monitoring only. Wikibusines: AI visibility agencies 2026.

What a useful audit must include. A credible AI visibility audit is not a single automated score. It should contain: (1) a defined prompt set of 10 to 25 questions your customers actually ask, (2) testing across at least three engines (ChatGPT Search, Perplexity, Google AI Overviews), (3) a competitor benchmark on the same prompts, (4) a citation-by-citation record with raw responses saved, (5) a prioritised fix list tied to specific pages, and (6) a re-test date so you can measure change. An audit that delivers only a score without the underlying prompt evidence is not actionable.

Ranklore's position. We offer a one-time AI Workflow Audit at S$750, which includes a 15-prompt baseline across three engines, a competitor comparison, a page-by-page fix list and a 45-minute walkthrough. This is our published price, not a claim that it is the market rate. It sits at the lower end of the international one-time audit band because we are a new boutique studio, and it is designed as an entry point rather than a full retainer.

Red flags. If an agency quotes a monthly retainer without first doing a baseline audit, you cannot measure whether the retainer is working. If an audit does not save the raw AI responses, you cannot verify its claims. If the price is below S$500 for a human-reviewed audit, it is likely an automated tool output with no analysis.

How to read the new open standards and free tools

Between September and October 2026, the "AI visibility audit" and "how to measure AI visibility" results shifted from agency PDFs toward an open scoring standard and free checkers. These are useful signals and a common source of confusion. Two developments matter.

The AI Visibility Score (AVS). In April 2026, a preprint on Zenodo proposed AVS, an open, deterministic 0-100 score combining 30+ SEO checks and 32+ AEO checks into one number. The authors submitted 155 queries to OpenAI's web search API, collected 816 citations and scanned 721 cited URLs. Median cited site scored 77 (Grade B); recommendation queries cited higher-scoring sites (mean 80.2) than local queries (mean 60.0); and the SEO-AEO gap was 16 points (80.6 vs 64.5), which the authors interpret as most sites not yet optimised for AI search. Xie, M. (2026), AI Visibility Score, Zenodo.

What is useful: the score is deterministic and open, so a site can be re-scored by the same checklist and the result reproduced. That is a genuine step beyond a proprietary "made-up rank" because the method is inspectable. What it does not do: AVS is a site-readiness checklist, not a measurement of what appeared in real answers. A page can score 90 on its checks and still be absent from a ChatGPT answer, because the checks measure crawlability, structure and content features, not observed citations. Treat AVS as a reproducible technical baseline, not as a substitute for a citation register.

Free and freemium checkers. Semrush's free AI Search Visibility Checker runs a 0-100 AI Visibility Score plus mention and citation counts across ChatGPT, Gemini, AI Overviews and AI Mode, free three times per day, drawing on a claimed 26M+ prompt and response database. Semrush free AI visibility tool, updated 14 October 2026. Wellows, a tracker with a free tier and paid plans from about US$37 per domain per month, audits ChatGPT, Gemini, AI Overviews, AI Mode and Perplexity (five surfaces at its Starter tier) and returns a citation score and per-platform coverage. Wellows pricing via CheckThat.ai, September 2026; DataAlly tool review, August 2026. In October 2026 Wellows also published a five-platform manual audit method, but that article's URL was not verifiable at the time of writing, so it is not cited here. Free tools like isvisible.ai and botric.ai similarly return a score or checklist on demand.

What is useful: a free checker gives you a fast baseline and a per-platform split for zero cost. What to watch: the score is computed by a vendor you cannot inspect, the prompt set is usually hidden, and a score alone contains no raw answers to verify. Our worked example above still applies: a number without the saved prompt, response and source URL cannot support a claim about market visibility. If a tool gives you a score, ask whether it exports the underlying answers; if it cannot, keep your own register for the decisions that matter.

Measurement is becoming a research field

By late October 2026, "how to measure AI visibility" returns more than agency methodology. The results now include an industry-body white paper, academic preprints and practitioner tool roundups. Three layers are forming, and each raises what a credible audit should cite.

The industry layer. In August 2026, the IAB published "Measuring Visibility in the AI Era", an industry-body attempt to define a common vocabulary: mention rate, citation rate and post-citation click-through rate, with disclosure requirements for measurement providers. It is a standardisation signal, not a settled standard; the report itself requires providers to disclose their method. IAB white paper, August 2026. The practical takeaway: when a vendor shows you a metric, ask which definition it uses and how it computed the denominator.

The adoption gap. Marketing-Interactive's 1 October 2026 industry review cites a McKinsey finding that only 16% of Fortune 500 consumer brands systematically track AI search performance, and argues Southeast Asia adoption is lower. Marketing-Interactive, 1 October 2026. Most vendors are measuring more rigorously than most buyers are paying for. That gap is why we recommend keeping a register even when you buy a tool: it puts you in the minority that can verify a claim.

The academic layer. Two 2026 preprints treat visibility as a measurement problem with defined statistics, not a dashboard number. "Quantifying Uncertainty in AI Visibility" builds a statistical framework for generative-search measurement, defining visibility metrics over sampled responses. arXiv 2603.08924. "From Stochastic to Stable: Rank Stability and Structural Sufficiency in AI Visibility Measurement" shows citation distributions in generative engines follow a power-law form and defines when a sample is large enough to report rankings at all. arXiv 2607.10341v2. The practical point for a buyer: answer orderings are stochastic, so one screenshot is a draw, not a measurement; you need a sample sized until the ordering stabilises. Backlinko's 8 October 2026 roundup of LLM tracking tools and Obsurfable's July 2026 share-of-voice framework sit in the same practitioner layer. Backlinko tool roundup, 8 October 2026; Obsurfable white paper, July 2026.

None of these layers changes the register method in this guide: a defined prompt set, repeated checks, raw answers saved. They raise the bar for what a credible audit must cite, which is exactly what the checklist in the audit section above covers.

GEO for Singapore SMEs: what actually moves the needle

"GEO" (Generative Engine Optimization) is now a standard line item in Singapore agency proposals. MediaOne positions GEO as a core service alongside SEO, with a dedicated service page updated October 2026 and an in-house tracker called "Digimetrics" measuring "Answer Share." MediaOne GEO service page, October 2026. Hashmeta includes AI search optimisation (GEO) in its Professional and Enterprise SEO plans. Hashmeta SEO service Singapore, August 2026. First Page now includes GEO/AEO as a default component of enterprise engagements. First Page SEO pricing, April 2026.

Terminology is not yet settled in Singapore. One agency ranking page still defines GEO as "Geographic Engine Optimisation", while the same provider's service page and most industry writing use "Generative Engine Optimisation". MediaOne GEO ranking guide, September 2026; lin.com.sg GEO category, October 2026. When comparing proposals, check which definition each side uses before comparing numbers.

The marketing is ahead of the evidence. A practical buyer should separate tactics with documented support from tactics that are currently claims.

What a Singapore SME should do first. Three actions have the strongest evidence-to-cost ratio:

1. Fix crawlability and indexability. If Google Search Console shows coverage errors, those pages are invisible to both search and AI answers. This is the foundation.
2. Write answer-first paragraphs. For each service page, include a one-sentence definition, a comparison table and a clear "who this is for" statement. Models extract these structures more reliably than narrative prose.
3. Sync entity data. Ensure your business name, address, phone number and service list are identical across your website, Google Business Profile and any directory listings. Inconsistent data causes models to treat them as different businesses.

What to defer. llms.txt, GEA/paid AI ads and proprietary "AI optimisation" add-ons are not yet worth a separate budget line for most SMEs. If an agency proposes GEO as an add-on costing more than the SEO retainer, ask which specific task in the table above it performs that the SEO work does not already cover.

This assessment is our interpretation based on available evidence as of October 2026. The field is changing quickly; provider documentation and controlled studies may alter these assessments. We will update this section when evidence changes.

Does your site need an llms.txt file?

llms.txt is a plain-text file placed at your domain root that describes your site's important pages for AI crawlers. It was proposed in late 2024 as a way to help AI systems understand your content structure. The question for most businesses is not whether the file exists, but whether it changes anything.

Google's position. On 15 June 2026, Google updated its official AI optimisation guide to state explicitly that llms.txt files are not needed for Google Search and neither help nor hurt rankings. Google's generative AI features are grounded in the same Search index, not a separate AI crawl path. Google's clarification on llms.txt, June 2026; Aergos summary, June 2026.

Other providers. OpenAI, Anthropic and Perplexity have not formally adopted llms.txt as a required or ranking-relevant file. Their crawlers primarily use robots.txt for access control and standard HTML for content retrieval. Some AI coding assistants and documentation tools may read llms.txt to understand a project's structure, but that is a developer-tooling use case, not a consumer-search visibility one. Coronium.io analysis of crawler adoption, May 2026.

The confusion. Chrome's Lighthouse tool began auditing for llms.txt in May 2026, which made it look like a Google requirement. It is not. Lighthouse audits many things that are not ranking factors. PikaSEO analysis of the Lighthouse contradiction, June 2026.

Practical recommendation. Adding a basic llms.txt is low effort, a 10-line Markdown file listing your key pages, and it does no harm. But it should not be prioritised over crawlability, indexability, content quality and structured data. If your site already has a clear sitemap.xml and robots.txt, llms.txt adds marginal value at best. If you are choosing between fixing a broken canonical tag and writing llms.txt, fix the canonical.

What can Google report directly?

Google announced dedicated generative-AI performance reports in June 2026 and noted worldwide rollout on 31 August. These report impressions and supporting breakdowns for Google's AI features. The data also appears in the overall performance report, so do not add it to the total a second time. Google's report announcement.

Keep that provider-reported view separate from third-party prompt experiments. If your reporting integration cannot retrieve the dedicated report, show that limitation instead of inventing a substitute.

Where do enquiries fit?

Track a separate enquiry record with landing page, referrer or campaign tags when available, customer-reported discovery and qualification outcome. Missing attribution stays unknown. An AI citation is evidence of appearance in an answer, not proof that it caused a sale.

What should you ask an agency to show?

Open one observation behind the chart. Check the saved question and answer, follow the citation, and compare the date and method with the previous period. Then ask which practical decision the result supports. If the agency cannot open the underlying evidence, the chart is not ready to guide spending.

Related field notes