More claims, not yet under a principle

Do not judge AI visibility from one check. Ask the same question daily and count on how many of the last 7 days your site was cited; expect brand mentions to be steadier than links.

When verifying Claude-User, check the bot IP list first. If an IP fails it but falls in Anthropic's published outbound range, label it 'published range', neither 'verified' nor 'fake'.

If your site has a public API, publish /.well-known/api-catalog as application/linkset+json, linking to each API's OpenAPI file and documentation. If you have no public API, skip it.

Before buying an AI-visibility tool, ask what it measures: human visits from AI apps, answer snapshots, or bot requests in your own logs. If your site has few pages, do not expect page-group split tests to work.

If you use noarchive or nocache to keep pages out of Bing's AI answers, add <meta name="Bingbot" content="ai-grounding:disallow"> or the matching X-Robots-Tag header, and record the setting in Bing Webmaster Tools before 2027-03-01.

Use Crawl Control to slow Bingbot during your busy business hours if server load is a problem. Do not raise the rate expecting new pages to be crawled faster.

To check indexing, use URL Inspection in Bing Webmaster Tools, or the url: operator on Bing, not the API's crawl dates. If you call the API, space calls about 6 seconds apart to avoid throttling.

If you post to Bluesky for your site: use an app password, reuse sessions instead of logging in for each post, add the bot self-label, use your domain as the handle via /.well-known/atproto-did, and never like or reply unless tagged.

Allow Googlebot in robots.txt on pages you want in Brave Search. Treat Bravebot lines in your logs as unverified. Check whether Brave has your pages on search.brave.com or through the Brave Search API.

Do not browse your own site with Web Discovery enabled to push it into Brave; one machine gives only 1 of the 20 views needed. Rely on real visitors from many different networks instead.

Do not buy ChatGPT Ads expecting the answer itself to name you; appearing in an ad is not being cited. When you monitor ChatGPT answers, count mentions inside ads separately from mentions in the answer.

Give every button, menu and form field on key pages a clear label (visible text or aria-label), plus role and state where needed. Run an accessibility checker to find unlabelled inputs.

If you run a read-only MCP tool, prepare the review package: legal and support URLs, test cases, a demo video, ja-JP text, and a clear statement of what your tool does that ChatGPT cannot do on its own.

Japanese shops: keep product pages crawlable and your own checkout working, since ChatGPT sends buyers to your site. Wait until OpenAI names Japan before building a product feed for the Agentic Commerce Protocol.

If your site is on Cloudflare, check whether Crawler Hints is turned on in your dashboard; it costs nothing on any plan. Keep submitting your important new URLs yourself as well.

If your site runs on a Cloudflare Worker with a Custom Domain, publish machine-readable files such as those under /.well-known/ with a small route Worker, separate from your page releases.

Do not expect AI click data in Search Console yet. Record your AI-feature impressions now as a baseline, and check Search Console again after 2026-12-03.

Make sure your site has a clear company profile page (who runs the company, parent company, contact details) and that it is in your sitemap. This is cheap whatever the true ratio is.

For questions nobody else answers, write one clear page with your own data, using the words people will ask with. Then get it indexed: submit it in both Google Search Console and Bing Webmaster Tools and confirm it shows as indexed.

When you buy or run AI-answer monitoring, ask how many times each question is asked, and treat a single answer as one sample, not as the result.

If you link your site from a GitHub README, add a tracking parameter to that link so your logs show whether anything arrives through it, and do not expect bots to follow it, since GitHub marks it nofollow.

When you repost an article on another platform, ask that platform to noindex the copy if you can; if you use a canonical link instead, point it only at a page you have confirmed is indexed in both Google and Bing.

Apply for the preview only if your service gives live answers or actions that Gemini or Search cannot find on their own, ideally with user sign-in; a read-only copy of your public data is not what Google asks for.

If you publish data, mark the page up with Dataset (name, a description of 50 to 5,000 characters, creator, license), keep it crawlable, list it in your sitemap, and allow several days for Google to find it.

On a page with no visible price or reviews, use Service markup instead of Product or SoftwareApplication, do not add invented offers or ratings, and then check the page with the Rich Results Test.

Bookmark each new or meaningfully updated page of your site once, from one account, with a one-sentence comment that is true to the page. Do not bulk-bookmark, run it on a dense schedule, ask others to bookmark, or use a second account.

If you run a Hatena Blog, take the API key from the blog's advanced settings (詳細設定) and post over HTTPS. Do not expect AI search bots to find a brand-new blog quickly; link it from places they already read.

Treat a Hatena Bookmark entry as an allowed path to your page, not a proven one. If you bookmark your own pages there, follow Hatena's spam rules, and check your logs for AI search bots arriving from it before investing more.

If you publish original data behind your pages, consider also publishing it as a Hugging Face dataset with a clear card (license, language, tags) linking to your source page, then check whether that link is nofollow.

If you want to refuse AI training, keep naming each training bot (GPTBot, ClaudeBot, Google-Extended, CCBot and others) in robots.txt now. You may add the common token too, but do not rely on it alone before vendors start honouring it.

Check all three layers: robots.txt, your server's response to each AI bot's user agent, and the X-Robots-Tag header. Review rental server, CDN and plugin settings for bot blocking you did not intend, then confirm with real bot visits in your logs.

Do not look for a Jev crawler in your logs or write robots.txt rules for it: Jev judges content that another tool has already fetched. Look for that fetching tool's user agent instead.

Treat llms.txt as optional and cheap. If you publish one, follow the format (H1 title, short summary, lists of links), but fix crawlable pages and your sitemap first, and do not expect AI answers to change because of it.

If you want to refuse AI training this way, add the __usage_ai-training token only as a forward-looking step: vendor support is planned through 2027 (Microsoft from 2027-02-01), and no top .jp site had written it as of 2026-11-02.

When choosing an AI visibility tool, include Japanese tools with real customers such as ミエルカGEO in your comparison, and use free trials before paying.

If you run a local shop, compare such services by what they measure: shop listings and AI recommendations for consumer keywords, not whether AI bots can read your own site.

Do not plan around this yet. Watch for an official OpenAI announcement, and keep your company and product information accurate and consistent on your own site meanwhile.

To verify Parallel's bots, check ShapBot against its published IP list, and check Web Bot Auth signatures against the key directories on shapbot.parallel.ai and shap-user.parallel.ai, since Shap-User has no IP list.

If you need a free Japanese press release channel, write a plain-text body of at least 300 characters and submit it through PR-FREE's form. Do not count on it for AI visibility until you see your release cited.

If you run a crawler or scanner, send a pull request adding its user agent (pattern, info URL, example) to these lists, so site owners' tools show its name instead of 'unknown' or 'Generic Bot'.

Fetch your robots.txt with a bot user agent and make sure it returns 200. Never use 401 or 403 on robots.txt to block or slow bots; write Disallow rules instead. On Amazon S3, a missing file can return 403.

Open the Search generative AI setting in Search Console for each property and its parent property, and keep it on Include if you want to appear in AI Overviews and AI Mode. Only property owners can change it.

Control ShapBot with the ShapBot token in robots.txt and check its visits against the published shapbot.json list. Do not expect to verify Shap-User by IP, because no list exists.

Put every fact in the visible page text first and let JSON-LD only repeat it. Do not hide extra facts for AI in FAQ markup, and do not add FAQPage expecting a rich result.

Post to TikTok by hand in the app, not by script. Put your link in the profile website field (business account or 1,000 followers), and label realistic AI-generated content.

Post automatically to X only through the API, with keys a person created. Show the automated account label linked to a human-run account, set a spending cap, and never automate likes, follows or replies.

Upload through the API with an OAuth token from a project set to In production, not by driving YouTube Studio in a browser. Unlock advanced features before counting link clicks, and disclose realistic synthetic content.

Use Zenodo for data you want permanently citable. Prepare the record with write-only access first, and have a person approve before publishing, because a published record cannot be deleted.