Search engineering,
end to end.
26 services across SEO, digital marketing and web development, run by the people who sign the work.
All services
+340%
organic patient growth for Dermacian, a dermatology clinic
Markets where
accuracy is the job.
Each market has its own review standards, page types and measurement limits. We plan around them from day one.
All industries
Healthcare & MedTech
Clinical review, accurate provider facts, proportionate claims.
FinTech
Fees, rates and the regulated entity stated plainly and dated.
B2B SaaS
Category, use-case, integration and security pages with distinct jobs.
Ecommerce & marketplaces
Rules for filters, stock states, sellers and reviews.
Legal & law firms
Jurisdiction, attorney involvement and advertising rules.
18 free tools.
No signup.
Can AI search quote your page? Paste a URL and the extractability checker scores it against nine structural checks.
All 18 toolsAI search 8
Technical 5
Local and global 2
Named people.
Published method.
Sakshi Kumari and Ashish Khan run MoxSEO's search work and sign the research they publish.
Meet the teamMoxSEO’s research methodology starts from a falsifiable question and a defined sample, scores public signals with disclosed rules, separates measurement from interpretation, states its limits and corrects errors in public. It is written as 13 numbered clauses, so you can check any finding against the rule it was meant to follow.
Each clause states the rule. The note beside it shows the rule applied in one of our three published studies, with the real numbers.
1.1This standard covers every MoxSEO study, the datasets published with it, and the client results we report.
1.2No study proves a universal rule from a limited sample. Collection dates, tool behaviour, geography and page changes can all affect a later replication. Research identifies patterns; it does not replace validation on a specific website, market or customer journey.
2.1A study starts from a falsifiable question and a defined unit of analysis.
2.2The study states its population, its selection rules, its collection window, and the decision it is meant to inform.
2.3Methods, sample boundaries, scoring rules, review steps and limitations are documented before a finding is presented.
3.1Exclusions are made by stated rule, not case by case. The crawler study removed infrastructure, CDN and DNS domains by pattern, because raw Tranco ranks those hosts at the top and they serve no robots.txt.
3.2Drop-outs are reported, not replaced, and the yield is published next to the finding it qualifies.
3.3Where a sample is not random, the study says which way the bias most likely runs.
4.1We record observable, public technical and content signals.
4.2Records are kept separate from interpretation.
4.3An absent signal is not treated as evidence of intent, quality or causation.
4.4Collection parameters are published with the study, so the fetch can be repeated.
5.1Criteria are documented and applied identically to every unit in the sample.
5.2Raw observations are retained where publication rights allow.
5.3Weights are disclosed. Where no published standard exists, they are declared as our judgement rather than presented as one.
5.4Verdict rules are stated in full. In the crawler study, a crawler is blocked when Disallow: / applies with no Allow exception, partial when some paths are disallowed, and allowed otherwise.
6.1Ambiguous cases are reviewed and kept visible, not quietly resolved.
6.2When a collection pass produces implausible results, the pass is investigated before anything is written, and degraded data is discarded.
6.3Errors caught during a study are disclosed in its limitations and marked as corrected.
7.1A study separates what it measured, what it interprets, and what still needs testing. Each is labelled as such, so a reader can accept the measurement and still reject the interpretation.
A ratio of 1.95. Anthropic’s pair shows 74 against 39, a ratio of 1.90.
Two independent operators with nearly identical ratios make a considered policy more likely than noise. It is still a reading of the numbers, not a measurement.
robots.txt records stated policy, not enforcement, and the sample skews toward large, Western sites.
Source: MoxSEO robots.txt scan, 26 July 2026. Read the study
8.1Every study discloses the limitations that apply to it: sampling bias, collection-date effects, tool limitations, geography, page changes, missing data and alternative explanations.
8.2Limitations sit in the study itself, next to the method, not in a separate document.
9.1We publish definitions, collection notes, datasets or working tools wherever practical. All three studies so far have an open CSV.
ai-crawler-study-2026-07-26.csv845 rows, one per domain, verdicts for 14 crawlersCSVanswer-extractability-study-2026-07-26.csv506 rows, per-check results, paragraph lengths, heading and link countsCSVcity-market-study-2026-07-27.csv148 rows, released under CC BY 4.0CSV9.2Where a check can be run on any site, it is published as a free tool. The checks from the crawler study power the AI Crawler Access Checker; answer structure can be tested with the AI Answer Extractability Checker. All 18 free tools are listed together.
10.1Every study shows when it was first published and when it was last updated.
10.2Dataset files carry their collection date in the file name.
10.3An update adds newer information without rewriting historical observations.
11.1Material errors are fixed and recorded in the public corrections log.
11.2Changes are classified as one of three kinds:
11.3Each entry records the original statement, the corrected information, the date, the reason, and whether the conclusion changed.
11.4To report an error, send the page URL, the exact statement, why it may be wrong, and your evidence to editorial@moxseo.com.
12.1A client figure is reported as observed in a defined context, scope and measurement period. It is never presented as a forecast for another site.
12.2Every headline figure states what it does not mean, as well as what it does.
12.3No invented clients, testimonials or metrics. Illustrative material is labelled as illustrative.
12.4Reporting separates completed work, technical health, leading signals, qualified demand and commercial contribution, and states the limits of attribution. See the case studies.
13.1Every study names its author, with a byline that links to the author’s page.
13.2Every study shows its publication date, its last update, and a link to the corrections log.
13.3Ashish Khan authored the crawler access and city-market studies; Sakshi Kumari authored the answer extractability study. Both are on the team page.
Each one names its author, its sample, its yield and its limits, and links to the full dataset.
| Study | Sample and yield | Headline | Dataset |
|---|---|---|---|
| AI crawler access studyAshish Khan, 26 Jul 2026 | 506of 845 Tranco domains returned a robots.txt | 20.6%block at least one of 14 AI crawlers, 104 of 506 | 845-row CSV |
| AI answer extractability studySakshi Kumari, 26 Jul 2026 | 243of 506 homepages scored, 48% yield | 3%score 80 or above; the median is 45 of 100 | 506-row CSV |
| City market AI readiness studyAshish Khan, 27 Jul 2026 | 148businesses across 20 cities in seven countries, 6 to 9 per city | 16%declare an author and dates on their site | 148-row CSV |
It is the standard every MoxSEO study, dataset and reported client result follows, written as 13 numbered clauses: a falsifiable question and defined population, rule-based sampling, public observable data, consistent and disclosed scoring, review of ambiguous cases, findings split from interpretation, stated limitations, open data, versioning, public corrections, careful client reporting and named authors.
Exclusions follow a stated rule, not case-by-case judgement, and drop-outs are reported rather than replaced. The crawler study removed infrastructure, CDN and DNS domains from the Tranco list by pattern; 506 of 845 domains returned a robots.txt. Where a sample is not random, the study says which way the bias most likely runs.
No. Where no published standard exists, the weights are declared as MoxSEO's judgement. The extractability checks carry 110 raw points normalised to 100, and the raw per-check results are published so anyone can re-weight them.
A measured finding is what the data shows, such as 72 of 506 sites blocking GPTBot against 37 blocking OAI-SearchBot. An interpretation is a reading of it, such as site owners pricing training and citation access differently. Each study labels the two, plus what still needs testing, so you can accept one and reject the other.
Each study links an open CSV: 845 rows for the AI crawler access study, 506 for the answer extractability study and 148 for the city-market study, 1,499 rows in total. File names carry the collection date. Checks that can run on any site are also published as free tools.
Material errors are fixed and recorded in the public corrections log. Each entry is classed as a correction, clarification or update, and records the original statement, the corrected information, the date, the reason and whether the conclusion changed.
Send the page URL, the exact statement, why it may be wrong and your evidence to editorial@moxseo.com.
A free 30-minute call with Sakshi Kumari or Ashish Khan, the specialists who run MoxSEO and wrote these studies. Bring your site; they will show you where it sits against the same checks, and say what the numbers cannot tell you.
A call is not an engagement. Plans start at $699 a month, and we never guarantee rankings.