How KindScore works

KindScore rates things to go and do: tours, wildlife experiences, visits, classes, tastings and shows, including dinner cruises and dinner shows, and stays where the stay is itself the experience, such as a wildlife lodge or volunteering at a sanctuary. It doesn't rate restaurants, bars, shops or ordinary places to stay.

A KindScore is an automated estimate of the sustainability impact of a travel activity, evaluated from publicly available information at the time the rating was produced. It uses a stated method applied the same way to every activity we rate, consisting of targeted web searches and AI analysis.

It's intended to give you a quick and clear overview of the effects of an activity, for you to form your own judgement and to read the sources behind it if you want to dig deeper. It is not travel-booking advice, a safety certification, a legal finding or a guarantee.

A KindScore rating may be incomplete, and it can go out of date as an operator changes how it runs things or as new information comes to light. See “Ratings quality” below for more on this, and how to tell us if something looks wrong.

Our method is informed by the GSTC Tour Operator Standard, alongside independent accreditation bodies, regulators, and editorial and NGO sources cited on individual ratings.

Each aspect of a KindScore rating - people, animals and the planet - is researched and assessed independently against the points below. The tool discovers the most important findings (points) and sorts them into either the “Good” or the “Concerns” category, based on the impact they have. Findings that only relate to one aspect do not influence the others. This is to give you the full picture of an activity's impact, and to make sure major points in one area don't hide points in another.

Expand each aspect to see details.

These three aspects come from our founding motivation for building Be TravelKind. Read our story for more on our values in sustainable travel.

The only place where the three aspects are combined is in the overall KindScore rating, to give you one simple rating for quick comparison. Each aspect counts towards it in proportion to how much we found about it. If an activity barely affects animals, Animals counts for very little, and if most of what matters is about animals, Animals counts for most. We strongly encourage you to read the individual aspect ratings and the points behind them, to give yourself a rounded understanding of the activity's impact.

Every point is assigned a score from 1 to 5, based on how significant it is, and the strength of evidence behind it. A major, well-evidenced point from an independent source will score 4 or 5, while a minor, thinly sourced or operator-claimed point will score 1 or 2.

For each aspect, the positive points are added and the negative points are subtracted, and the resulting score lands in one of five bands, from Very Poor to Excellent.

The overall KindScore is an average of the three aspect scores, weighted by how much we found about each one, so the aspect an activity mainly affects has the biggest say. An aspect the activity doesn't affect at all has no say, so a museum can still be Excellent with nothing to report on animals.

A serious concern in any aspect still holds the overall back, however good the rest is: the overall can be at most one band better than that aspect, or two when the concern rests on thin evidence. And an activity built around animals can't rate above Good if we found nothing about how its animals are treated.

Some points are worth knowing, but can't reliably be given a score. This could be because the point is disputed, or because the impact depends on operating conditions we couldn't verify for the specific operator. Those are shown separately in the “Worth knowing” section to give you the information without affecting the score.

Every result has a “Show how this score was calculated” button that shows each point's scores and how they add up to the rating.

KindScore looks for evidence the same way for every activity. It runs web searches on the operator's own site, independent accreditation and regulatory bodies, news coverage, blogs, reviews, and other public sources (see “How we keep ratings as fair as possible” below).

Following this, it uses AI to read through the top results that search turns up, and to compile the individual, distinct points that make up a rating. Read more about our view on AI, including our stance on data center construction and further development of AI.

We acknowledge that not all relevant information can be found this way. Some search results don't surface well, some sites cannot be read automatically and some information is simply not available online.

Many tourism activities carry an expectation among the public of their impact, and how positive or negative it might be. This may not reflect the nuance of how an individual operator actually runs that activity. Therefore, KindScore looks for specific, evidenced facts about the operator's own practices to inform the assessment. That is why every KindScore rates one specific operator: when we can't tell which business runs an activity, we ask you to choose or name one, rather than rating the type of activity in general. For the same reason, when a company runs several different trips, we rate one trip at a time: each can treat people, animals and the planet quite differently.

Some activities cause harm in ways which are not justifiable by any operator-specific nuance, such as conduct that is illegal, causes severe suffering to people or wildlife, or causes harm on a systemic scale. This is always treated as serious, regardless of other findings about the operator.

Any finding specific to an individual operator must link to a real, public source. A small icon-marked search link appears only for a point that isn't specific to any one operator, such as general knowledge about an activity type.

Tap any point on a KindScore result to see where it came from, and to get a plain-English explanation of why it matters.

Each source is labelled with the kind of thing it is, worked out from its web address and from how the page describes itself:

• Official record: a government, regulator or court.
• Accreditation body: an organisation that sets standards and accredits or certifies operators, with no legal power over them.
• Independent publication: an edited publication, such as news, a magazine, research or an NGO.
• Operator statement: what the operator being rated says about itself, on its own site.
• Promotional content: listings and articles from a booking platform, directory, tourism board or sponsor, written or paid for by businesses with a stake in what they cover.
• Other operator: another tourism business's own site. The label says only that it isn't the operator being rated, not that the two are connected.
• Community content: reviews, social media, forums, wikis and personal blogs.

Where a page says when it was published, that date is shown with its source. It is the page's date, not necessarily the date of what it describes. A concern whose newest source is more than five years old shows that year beside it, so an old finding never reads as current.

We never judge an individual outlet's credibility or bias, and we keep no list of trusted publications. A finding is more convincing when independent sources agree on it, so no single outlet decides a rating. How much a source counts follows from its kind: a site we can't yet place counts for less than an official record or a page that declares itself edited news or research.

Some pages are not evidence at all, whatever they say: text written by an AI model with no human author or editorial process, pages that copy other sites wholesale, and KindScore's own ratings, since citing a rating as evidence for itself would be circular. New ratings skip them. Our own blog posts, which come from our first-hand research, remain sources like any other.

Every rating shows how well-evidenced it is as a whole. It is High when an official record, or edited news or research publications on more than one site, stand behind it, Moderate when it leans on the operator's own statements or a single independent source, and Limited when it rests on only a few findings, mostly on reviews, forums and social posts, or mostly on sources more than five years old. A rating with Limited evidence can't score Excellent or Very Poor: there isn't enough corroborated evidence for either extreme. It is worked out when the rating is made and doesn't change unless the activity is rated again.

KindScore reports what its sources found, in the plainest words that capture it. Every point shows what kind of source it comes from and links to it, so you can read the detail there. Where a finding implies an impact you might not spot, the point may say so, worded as a possibility rather than something the source said.

Allegations of serious wrongdoing, such as harm to people or animals, crime, exploitation or a breach of the law, appear only as the conclusion of the source that reached them, never in our own voice, and so does any major concern. An operator's own claims about how ethical or caring it is are shown as its claims. Words like “scam”, “fraud”, “illegal” or “abuse” carry specific meanings, and are only used when a source directly establishes that meaning: a conviction, a regulatory or court finding, a documented legal ban, or the operator's own admission.

People are referred to by their role, never by name, and past events are dated rather than presented as current practice. This way of writing never softens a finding: how much a point counts depends only on its evidence.

A KindScore rating can only draw on what's publicly available online at the time we run it. Some real, relevant information doesn't surface well in a search: it might be behind a login, or on a page search engines don't index well, inaccessible to code (e.g. in a PDF) or simply not exist on the internet. Occasionally a source we find may be outdated, or wrong.

A rating is an automated estimate from what KindScore could find. It is not a guarantee that all relevant information was found, or that everything found was current and correct. We encourage you to read the included sources and to form your own judgement.

While we'd like to ensure every search result is read in detail, AI has a real environmental and financial cost, and the most important information is almost always found in the first few results. We therefore limit the number of results analysed to a practical amount. We want to ensure we remain responsible with our use of AI and make serious efforts to reduce its usage without compromising the quality of our ratings, as detailed in Our View on AI.

You can help us improve inaccurate or out-of-date ratings. If you ran the rating yourself, you can tap “Disagree? Report this rating” underneath it at any time. If you found a rating another way, email us at hello@betravelkind.com with the rating's share link and what you believe is inaccurate or out of date, and we'll review it.

A rating is only as trustworthy as the process behind it:

• KindScore ratings are produced by a consistent, automated process so no bias on our part can skew the outcome per operator.
• Every activity goes through the same process, regardless of the operator's size, location, or reputation.
• Every search we run is built from the operator being rated and fixed search terms, so the same process is applied to every operator.
• To reduce bias, skew or hallucination in the results, our automated search terms and AI prompts avoid specifying examples of operators or practices.
• If you think a rating is wrong, we encourage you to report it. See “Ratings quality” above. We guarantee to personally review every report.
• We do not accept payment, gifts, free experiences, or any other incentive from an operator in exchange for a better rating.
• Before a rating page is listed in search engines, a person at Be TravelKind reviews it and decides whether it is listed. Ratings you run yourself in the tool appear straight away, without review.
• A person never sets a score or changes how much a finding counts. The only changes we make by hand are corrections to findings that are wrong, shown openly. See “Corrections” below.
• We sometimes write our own blog posts about an operator, based on our own research and experience. Where we have, that post can be used as a source for the KindScore rating, but it's treated as one source among many, and never entirely decides the outcome on its own.

When a report shows that a finding is wrong (it describes a different business, misreads its source, or better evidence contradicts it), we correct the rating rather than leave the error standing. A rating can be corrected in two ways:

• Removing a finding that is wrong. The score is recalculated without it, and the rating notes how many findings were removed and when.
• Rewording a finding, for example to add a date or to attribute a claim to its source. Its weight and source stay the same, and it is marked “Corrected” with the date.

We never change how much a finding counts or set a score by hand, and a correction is never offered in exchange for anything. Where a correction isn't enough, we run the rating again. We keep a record of every correction, but we don't republish the wording we removed.

If you run a business we've rated, or you think something in a rating page is wrong, use the “Is this your business?” link on that page. A person reviews every request and replies within 2 working days. While we check, we may take the page out of search engines and show that it is under review in place of the rating. A rating page that shouldn't stand is removed.

A rating page is reviewed again at least once a year. If it hasn't been, it stays visible but leaves search engines until we've checked a fresh rating.

Affiliate links, advertising and commercial partnerships do not influence KindScore ratings, evidence selection, alternatives or editorial conclusions.

We never sell ratings, changes to a rating, a place in suggested alternatives, or the removal of a concern. Payment, partnership or advertising never affects the outcome or the speed of a review.

When you ask for alternatives, we look first at activities we have already rated. One is only suggested if it is rated at least as well as the activity you are looking at, both overall and on that activity's weakest area, and never below Neutral. Among those, we favour the ones that are better rated, more similar in kind, and closer.

Where we have too few of those, we search the web for other operators. These are marked “Not yet rated”: they are suggestions to look into, not recommendations, because we have not assessed them. You can rate any of them yourself.

Sometimes we also suggest a gentler way to do the same thing without an operator at all. That is an idea, not a rating.