SolarBazaar

Legal

AI Transparency Statement

How SolarBazaar uses artificial intelligence — and, just as importantly, where we refuse to.

Effective date
Operated by
MAKSOF GLOBAL LTD
Governing law
England and Wales

SolarBazaar, operated by MAKSOF GLOBAL LTD, uses AI tools in parts of how we research, write and organise content. We think you are entitled to know where — so this statement describes it specifically rather than in general terms.

The short version: AI helps us draft, summarise, translate and structure information. It does not decide which products are good, it does not invent test results or reviews, and nothing reaches the Platform without a person reviewing it and being accountable for it.

If you find something wrong, section 8 explains how to report it. We read every report.

1. Our commitment to transparency

A lot of websites now use AI without saying so. We think that is the wrong approach, particularly in a field where readers are weighing up a purchase that will sit on their roof for twenty-five years.

We publish this statement so you can judge our content on an informed basis: what was machine-assisted, what was checked by a person, and which parts of our work we deliberately keep away from automation. If you know how a page was made, you can decide how much to rely on it.

Our position is straightforward. AI helps us cover more ground, faster, in more languages. It does not get to decide what is good, and it does not get to make claims we have not verified.

2. How we use AI

We use third-party AI models — principally large language models — in the following parts of our workflow.

  • Drafting and structuring explanatory content. For general educational articles (how a string inverter differs from microinverters, what a temperature coefficient means), AI produces a first draft or an outline that an editor then rewrites, corrects and expands.
  • Summarising long documents. Manufacturer datasheets, installation manuals, standards documents and industry reports are long and repetitive. We use AI to condense them into working summaries, which are then checked against the original before anything derived from them is published.
  • Normalising and classifying product data. Manufacturers publish specifications in inconsistent units, formats and terminology. AI helps map those into our common schema — assigning a product to a category, extracting a rated power or efficiency figure, matching a variant to a product family — so that comparison tables line up.
  • Translation and localisation. We use AI to translate content between languages and adapt it for regional audiences, including terminology that differs between markets.
  • Meta descriptions, headlines and page summaries. Short-form supporting copy such as search snippets, titles and card summaries is frequently AI-generated from the underlying article.
  • Search and recommendations. AI-assisted techniques help interpret search queries, match them to relevant products and articles, and rank results by relevance.
  • Suggesting related content. Links to related articles, products and calculators are generated by automated similarity matching rather than curated by hand on every page.

3. Where we do not use AI

These limits matter more than the list above, so we state them plainly.

  • AI does not set editorial verdicts, scores or rankings. Where we say one product is a better choice than another, or place products in an order of preference, that judgement is made by people against stated criteria. A model may help assemble the underlying data; it does not decide the conclusion.
  • AI is not used to fabricate test results, measurements or reviews. We do not publish performance figures, test outcomes or hands-on impressions that were generated rather than obtained. If we have not measured something ourselves, we say where the figure came from instead of inventing one.
  • AI does not generate user reviews, ratings or testimonials. Any review attributed to a person is from a person. We do not synthesise customer feedback, pad review counts, or use models to write quotes attributed to real or invented individuals.
  • AI does not make safety or certification claims about a named product without a cited source. Statements about compliance marks, certifications, standards conformance, fire ratings or safety recalls are only published when traced to the certifying body, the regulator or the manufacturer's own documentation.
  • AI does not silently change published facts or figures. Corrections to specifications, tariffs or prices go through the same human review as the original.

4. Human review process

Nothing reaches the Platform on the strength of a model's output alone. Our working process is as follows.

  • A human editor reviews every AI-assisted draft before publication. Review means reading the whole piece for accuracy, structure and tone — not skimming it.
  • Technical claims and figures are checked against primary sources.Product specifications go back to the manufacturer's current datasheet. Standards and safety claims go back to the relevant standards body or certification listing. Tariff, incentive and net-metering claims go back to the responsible government department, regulator or network operator. Where a claim cannot be traced to a source, it is removed or rewritten as an explicit uncertainty.
  • Numbers are treated as high-risk. Figures are the output most likely to be confidently wrong, so quantities, units, dates and currency values are checked individually rather than accepted with the surrounding prose.
  • Calculator formulae are written and reviewed by people. The methodology behind each calculator, and its assumptions, are authored and verified by our team. AI is not used to invent a formula.
  • A named human is accountable for each published page. Every article and product record has an identified person responsible for its content, who signs it off and handles corrections to it. Accountability does not transfer to a tool.

5. Content labelling

Where AI has played a material part in producing a page, we indicate it on the page rather than only in this statement.

  • Articles that began as an AI-assisted draft carry a note indicating AI assistance alongside the human editor responsible for the final text.
  • Product records built from automated extraction of manufacturer documentation indicate that their specifications are compiled from manufacturer sources, and link to the source document where one is publicly available.
  • Machine-translated content is identified as such, with the original language noted.
  • Content written entirely by a person without AI assistance carries no AI label.

We do not label short supporting copy such as meta descriptions and card summaries individually — this statement covers them. We also do not treat routine tooling (spelling and grammar checking, formatting, code assistance) as AI-generated content.

6. Possible inaccuracies and known limitations

AI review reduces errors. It does not eliminate them, and the failure modes are specific enough to be worth naming.

  • Fluent but incorrect output.Language models produce confident, well-formed text regardless of whether the underlying claim is true. A fabricated specification reads exactly like a correct one — commonly called "hallucination". This is why our review focuses on verifying claims rather than on how a passage reads.
  • Stale training data. Models are trained up to a cutoff date and may reproduce superseded product ranges, discontinued models, withdrawn schemes or historic prices as though they were current.
  • Missed regulatory and tariff changes. Incentives, export rates and grid rules change frequently and often at short notice. Both models and published pages can lag behind a change by weeks or longer.
  • Regional conflation.Models can blend requirements from different markets — applying one country's wiring rules, certification marks or subsidy structure to another.
  • Unit and conversion errors. Mistakes between kW and kWh, W and Wp, or between metric and imperial units are a recurring risk in extracted specification data.
  • Lost nuance in summarisation. Condensing a datasheet can drop the conditions a figure depends on — test conditions, temperature, tolerance ranges — leaving a number that is technically present but misleading in isolation.

Verify anything you intend to rely on for a purchase or installation decision. Check specifications against the manufacturer's current datasheet, confirm incentives and tariffs with the responsible authority or your utility, and get a written quotation from a qualified installer who has surveyed your site. Please also read our disclaimer.

7. Source data and attribution

Our factual content is built on identifiable sources.

  • Product specifications come from manufacturer datasheets, installation manuals, product pages and technical bulletins.
  • Certification and standards information comes from standards bodies, certification and test houses, and published conformance listings.
  • Incentives, tariffs and grid rules come from government departments, energy regulators, and distribution network operators.
  • Market and pricing context comes from published industry reports, trade associations and reputable trade press.
  • News is attributed to the originating outlet, with a link to the original report.

We cite sources for factual claims and link to primary documents where they are publicly accessible, so you can check our work rather than take it on trust. Where a source is behind a paywall or not publicly available, we name it even if we cannot link to it. Where a figure is our own estimate or model output, we say so instead of implying an external source.

8. How to report an error

If something on SolarBazaar looks wrong, please tell us. Reports from readers, installers and manufacturers are one of the most effective checks we have, and we would much rather hear about a mistake than leave it in place.

Where to send it

What to include

  • the URL of the page;
  • the exact statement, figure or specification you believe is incorrect;
  • what the correct information is;
  • a source we can verify it against — a manufacturer datasheet, a standards or certification document, or an official government, regulator or utility page.

What we do with it

We review every report we receive. Where we confirm an error, we correct the page or annotate it. Where we cannot confirm a report, we will say so rather than quietly leave it unresolved. You do not need to give a reason for reporting something, and you do not need to be a customer, reader or industry professional to do it.

9. Corrections policy

We distinguish between routine maintenance and material corrections.

  • Routine updates— refreshing a price, adding a newly released model, tidying wording — are made without a notice, and the page's last-updated date changes.
  • Material corrections— where a factual claim, specification, figure or conclusion was wrong in a way that could have affected a reader's decision — are noted on the page itself, stating what was corrected and when. We do not delete the fact that a correction happened.
  • Withdrawn content — where a piece cannot be corrected, we remove or replace it and explain why on the page where practical.

This applies equally to AI-assisted and entirely human-written content. How a page was produced does not change our obligation to fix it.

10. Editorial independence

Using AI does not change the separation between our commercial and editorial work.

  • Advertisers and sponsors have no influence over our editorial assessments, comparison criteria, scores or rankings — and no influence over how AI tools are configured or applied.
  • We do not accept payment to have a product described more favourably, to remove a criticism, or to exclude a competitor.
  • Advertising, sponsored placements and affiliate relationships are disclosed and labelled.
  • Automated ranking and recommendation systems are tuned for relevance to the reader, not to favour commercial partners. Paid placements are labelled and shown as distinct from organic results.

The full position is set out in our advertising policy.

11. Data used for AI training

Your data

We do not submit personal data to third-party AI model providers for the purpose of training their models. Account details, contact-form messages, calculator inputs and saved items are not sent to model providers as training material.

Where a feature involves sending content to a third-party model in order to operate — for example generating a summary — we send only what the task requires, and we use providers under terms that do not permit our submitted content to be used to train their general-purpose models. How we handle personal data generally, including the legal bases we rely on and your rights, is set out in our privacy policy. Data-protection enquiries go to info@solarbazaar.io.

Our published content

Our public pages are a different matter. We allow named AI crawlers to access our published content — our robots directives explicitly permit the major AI search and answer crawlers alongside conventional search engines. This is a deliberate choice: we want solar information to be available to people wherever they ask the question, including inside AI assistants.

Allowing crawler access is not a waiver of our rights. Our content remains the property of MAKSOF GLOBAL LTD and is subject to the intellectual property and prohibited-conduct provisions of our terms and conditions, which prohibit bulk extraction and redistribution of our product database. We may change which crawlers are permitted at any time.

12. Governance and review

AI tooling changes quickly, and a transparency statement is only useful if it keeps describing what we actually do.

  • We review this statement whenever we materially change how AI is used on the Platform — a new tool in the workflow, a new content type, or a change to our review process.
  • We review it periodically even when nothing has changed, to confirm it is still accurate.
  • The effective date at the top of this page reflects the current version.
  • Human accountability for published content is a fixed requirement. Changes to tooling do not change who is answerable for a page.

If our practice ever diverges from what is written here, this statement is wrong and we want to know — the reporting route in section 8 applies to this page too.

13. Contact

SolarBazaar is operated by MAKSOF GLOBAL LTD, registered in England and Wales.

You can also reach us through the contact form.