The Complete AI Visibility Playbook: Copilot, Entities, Measurement and AI Agents
The final part of the AI Search Optimization series: how Bing powers Copilot, how AI decides who to trust, how to measure it all, and a 90-day plan.
This is the final part of a five-part series on optimizing your website for AI search. Part 1 covers every platform; Part 2 covers Google; Part 3 covers ChatGPT; Part 4 covers Claude, Perplexity and the other assistants.
Over the last four parts, we’ve looked at how each AI platform finds, reads and cites websites. This final part brings it together.
It covers four things the series hasn’t tackled in depth yet:
- Microsoft Copilot and Bing, the one AI platform that shows you exactly where you’re cited.
- Entities and brand mentions: how AI decides which companies to trust.
- Measuring AI visibility across every platform.
- AI agents: the assistants that don’t just read your website, but use it.
It ends with a 90-day plan you can follow, whatever kind of website you run.
Microsoft Copilot and Bing
Why Bing matters more than you think
Microsoft says its family of Copilot assistants passed 150 million monthly active users in October 2025. Copilot answers are built on Bing’s search index. Microsoft describes Copilot as “powered by Bing’s search index.”
That makes Bing much more than “the other search engine.” If your site isn’t well indexed in Bing, it’s very unlikely to be cited by Copilot. Many site owners have never even verified their site in Bing Webmaster Tools.
Microsoft doesn’t document a separate Copilot crawler. Its controls all work through Bing: bingbot for crawling, and Bing’s robots meta tags for how content is used.
What Microsoft says gets content into AI answers
In October 2025, Microsoft published a guide to getting content included in AI search answers. It’s one of the most practical pieces of official AEO guidance from any platform:
- The basics still matter. “Crawlability, metadata, internal linking, and backlinks remain essential.”
- Clear titles and headings. Page titles should “clearly summarize what the content delivers,” and the H1 should match the title. Headings “act like chapter titles that define clear content slices.”
- Questions and answers. “Direct questions with clear answers mirror the way people search.”
- Lists and tables are “especially effective for how-to queries and feature comparisons.”
- Specific facts. “Anchor claims in measurable facts.” Microsoft’s example: “42 dB dishwasher designed for open-concept kitchens” instead of “quiet dishwasher.”
- Self-contained sentences that “make sense even when pulled out of context.”
And what to avoid:
- Hiding answers in tabs or expandable menus: “AI systems may not render hidden content.”
- Putting core information only in PDFs (“use HTML”) or only in images.
- Long, unbroken blocks of text.
- Big claims like “next-gen” with nothing to back them up.
IndexNow: tell Bing the moment something changes
AI answers are only as fresh as the index behind them. IndexNow lets your site notify search engines the instant a page is added, updated or deleted, instead of waiting to be recrawled.
How it works, in plain terms:
- Create a key (a string of letters and numbers) and host it as a small text file on your site.
- Whenever a page changes, send its URL and your key to an IndexNow endpoint. You can send up to 10,000 URLs at a time.
- Participating search engines share the notification with each other.
Bing, Yandex, Seznam, Naver and others take part. Google doesn’t. Many platforms and plugins (including popular WordPress SEO plugins and Cloudflare) can send IndexNow notifications automatically.
The AI Performance report
In February 2026, Bing Webmaster Tools launched an AI Performance report in public preview. It’s the first report from any major platform that shows actual citations:
- Total citations: how many times your pages appeared as sources in AI answers across Copilot, Bing’s AI summaries and some partner services.
- Average cited pages: how many of your pages are cited per day.
- Grounding queries: the search phrases the AI used when it found and cited your content.
- Page-level activity and trends over time.
In March 2026, Microsoft added a view that maps grounding queries to the pages cited for them.
Two caveats from Microsoft itself: the numbers are a sample of citation activity, and they show how often you’re cited, “not page importance, ranking, or placement.”
Bing’s content controls
noarchive: Microsoft said in 2023 that content with this tag won't be included in Bing Chat (now Copilot) answers and won't be used for training.nocache: limits AI answers to your URL, title and snippet.
Use them deliberately; they reduce your visibility in Copilot.
Entities and brand mentions: how AI decides who to trust
Every AI assistant has to answer a hidden question before it can answer yours: which company, product or person is this about?
There are many businesses called “Apex,” “Summit” or “Nova.” There are products with the same name in different industries. AI systems resolve this by building a picture of each entity, a distinct real-world thing, from everything they read about it.
If your website, LinkedIn page, review profiles and press coverage all describe you the same way, that picture is sharp, and you’re easy to recommend with confidence. If they disagree, the picture blurs. Assistants may confuse you with someone else, describe you with outdated facts, or leave you out.
Step 1: Write a canonical fact sheet
Before you change anything, decide what’s true. Write one document with your official:
- company name (and any former names)
- founding year and headquarters
- what you do, in one clear sentence
- product and service names, each with a one-line description
- leadership names and titles
- key numbers you’re happy to publish (customers, locations, years in business)
This becomes the reference for every profile and page.
Step 2: Make your own website unambiguous
- An About page that states those facts plainly.
- Organization structured data on your homepage or About page. Google’s documentation (updated September 2026) explains that the
sameAsproperty links to your profiles on other websites, such as social media or review sites, and that some properties are used "behind the scenes to disambiguate your organization from other organizations." - Author and leadership pages with real names, roles and expertise, linked from the content they write.
- The same names everywhere on the site: don’t call a product three different things on three pages.
Step 3: Fix the profiles AI reads
Check and align your facts on the places AI systems draw from: LinkedIn, Crunchbase, Wikidata, Google Business Profile, industry directories and review sites such as G2 or Capterra (for software) or Trustpilot (for consumer brands). If you have a Wikipedia article, make sure it’s accurate, following Wikipedia’s rules on conflicts of interest.
As we saw earlier in the series, independent research has found Wikipedia to be ChatGPT’s most-cited source and Reddit to be Perplexity’s. These sites shape how AI describes you.
Step 4: Earn mentions, don’t manufacture them
AI assistants weigh what others say about you: coverage in respected publications, genuine community discussions, reviews, podcast and conference appearances. Google warns that chasing inauthentic mentions “isn’t as helpful as it might seem,” and its spam policies cover content made mainly to manipulate AI responses.
The honest route works better anyway: publish original research others want to cite, contribute expert commentary, and take part in the communities where your customers ask questions.
Step 5: Check how AI describes you
Ask ChatGPT, Gemini, Claude, Perplexity and Copilot: “What is [your company]?” and “Who are the main providers of [your category]?” Note every wrong fact, then trace it back to the source that’s spreading it.
Measuring AI visibility across every platform
You can’t improve what you can’t see. Here’s a practical measurement stack, from the most precise data to the roughest.
Layer 1: Official platform reports
Only two platforms give site owners real AI data today:
- Google Search Console, Generative AI performance report: impressions in AI Overviews and AI Mode, by page, country, date and device. No clicks, and no Gemini app. (Covered in Part 2.)
- Bing Webmaster Tools, AI Performance report: citations in Copilot and Bing’s AI summaries, cited pages and grounding queries. (Covered above.)
Check both monthly. Compare the pages that appear with your most important pages.
Layer 2: Referral traffic
In Google Analytics 4 or your analytics tool, create a custom channel group called “AI assistants” that includes:
- chatgpt.com (ChatGPT also adds
utm_source=chatgpt.com) - perplexity.ai
- copilot.microsoft.com
- gemini.google.com
- claude.ai
Track sessions, landing pages and conversions from this channel. AI referral traffic is often small, but worth watching closely because it comes from people who’ve already asked a specific question.
Layer 3: Server logs
Count visits from the search and citation crawlers: Googlebot, Bingbot, OAI-SearchBot, Claude-SearchBot, PerplexityBot, Meta-WebIndexer, Applebot and Amzn-SearchBot. Also count user-triggered fetchers like ChatGPT-User, Claude-User and Perplexity-User. These show when a person asked an AI to read your page.
Get Harish Ratanpara’s stories in your inbox
Join Medium for free to get updates from this writer.
If an important page never gets a visit from a search crawler, it can’t be cited.
Layer 4: Prompt testing
Write 20 to 30 questions your customers really ask, including comparison and “best provider” questions. Each month, run them in ChatGPT, Gemini, Copilot, Claude and Perplexity, and record:
- Were you mentioned?
- Were you cited with a link? Which page?
- Was the description accurate?
- Which competitors appeared?
A simple spreadsheet is enough. AI answers vary from one run to the next, so look at trends over months, not single answers.
A word on third-party tools
Many tools now promise to track “AI rankings.” They can help with prompt testing at scale, but treat their numbers as samples. Google states plainly that no third-party tool has access to its internal ranking or AI systems.
AI agents: when AI doesn’t just read your site, it uses it
So far, this series has been about AI that reads your website and cites it. The next shift is AI that uses your website: opening pages, filling in forms, comparing products and completing tasks on someone’s behalf.
This is already happening:
- Google-Agent is used by “agents hosted on Google infrastructure to navigate the web and perform actions upon user request.”
- ChatGPT agent can browse and complete tasks for users.
- Claude in Chrome lets Claude take actions on websites inside a user’s own browser.
- Perplexity’s Comet browser has an assistant that acts on the pages people visit.
Two things to know about these agents. First, because a person asked for the action, Google says its user-triggered fetchers “generally ignore robots.txt rules”; other companies say similar things about theirs. Second, many agents “see” your site through screenshots, the page’s code and the accessibility tree, the same structure screen readers use.
Make your site easy for agents (and people) to use
Google’s web.dev guide to agent-friendly websites (updated April 2026) gives clear, practical advice:
- Use real
<button>and<a>elements, not clickable<div>or<span>elements. - Connect every form label to its field with the
forattribute. - Keep interactive elements visible and big enough to target.
- Avoid transparent overlays that cover buttons.
- Keep layouts stable, so elements don’t jump around while an agent is looking at them.
OpenAI’s publisher guidance makes the same point: follow accessibility best practice with descriptive roles, labels and states on buttons, menus and forms.
As Google’s guide puts it: “Everything we suggest to make a site ‘agent-ready’ also makes sites better for humans.” Accessibility work now pays off twice.
New standards to watch
These are early, so treat them as things to watch rather than things to rush:
- WebMCP: a proposal, developed in a W3C community group with editors from Microsoft and Google, that lets a website tell AI agents which actions it supports (“search products,” “book a demo”) instead of agents guessing from the layout. Chrome began an origin trial in version 149 in mid-2026. It isn’t a W3C standard yet.
- Lighthouse Agentic Browsing: an experimental Lighthouse category that checks agent-readiness: WebMCP tools, agent accessibility, layout stability and whether you have an llms.txt file. It reports a pass ratio, not a score.
- Universal Commerce Protocol (UCP): an open standard Google introduced in January 2026 with Shopify, Etsy, Wayfair, Target, Walmart and others, so AI agents can complete purchases. Its first uses are checkout in AI Mode and the Gemini app, through Merchant Center.
- Signed agents (Web Bot Auth): a way for AI agents to prove who they are with cryptographic signatures, so you can tell a real assistant from a fake one. OpenAI signs ChatGPT agent requests this way, and Google says it’s experimenting with it.
What to do now
- Run an accessibility audit and fix unlabeled buttons, forms and menus.
- Make key actions (contact, demo booking, checkout, search) simple, standard HTML.
- Check your bot protection: make sure it can tell verified agents apart from abusive scrapers, rather than blocking everything automated.
- If you run ecommerce, keep Merchant Center and product feeds accurate; agentic checkout builds on them.
- Keep an eye on WebMCP, but don’t rebuild your site for it yet.
Your 90-day AI visibility plan
Here’s everything from the series turned into a plan you can follow, whatever kind of website you run.
Days 1–30: Access and measurement
- Audit robots.txt, your CDN and your firewall. Allow every search and citation crawler: Googlebot, Bingbot, OAI-SearchBot, Claude-SearchBot, PerplexityBot, Meta-WebIndexer, Applebot, Amzn-SearchBot.
- Decide your training policy (GPTBot, ClaudeBot, Google-Extended, Applebot-Extended and others) as a separate business decision.
- Check your Search Console generative AI setting.
- Verify your site in Bing Webmaster Tools and turn on IndexNow.
- Make sure key content is in plain HTML, not hidden behind JavaScript, tabs or PDFs.
- Set up the “AI assistants” channel in analytics.
- Write your list of 20 to 30 test prompts and record a baseline.
Days 31–60: Content and entities
- Write your canonical fact sheet and align your About page, structured data and external profiles to it.
- Rewrite your ten most important pages so each heading is a real question and each answer comes first.
- Add specific facts, figures, examples and sources to those pages.
- Publish one piece of original information others will want to cite: a survey, benchmark or data study.
- Fix any wrong facts AI assistants repeat about you, at their source.
Days 61–90: Authority, agents and review
- Earn mentions in industry publications, communities and podcasts.
- Refresh review profiles and encourage genuine reviews.
- Run an accessibility audit and fix forms, buttons and menus for agents and people alike.
- Review Google’s Generative AI report and Bing’s AI Performance report.
- Re-run your test prompts and compare with your baseline.
- Decide what to repeat every quarter.
The whole series in five sentences
- Part 1: AEO and GEO build on SEO; access, content, entities and authority work on every platform.
- Part 2: Google’s AI features use the Search index and Search controls, while the Gemini app is governed by Google-Extended.
- Part 3: ChatGPT search depends on OAI-SearchBot, not GPTBot, and shopping depends on clean product data.
- Part 4: Claude, Perplexity and the other assistants each separate search bots from training bots, so check every one.
- Part 5: Bing powers Copilot and shows real citations, entities decide who AI trusts, and agents are the next visitors to plan for.
FAQ
Which AI platform should I prioritize? Start with Google and Bing, because they cover AI Overviews, AI Mode and Copilot, and give you real data. Then focus on whichever assistants your customers use most.
How long until I see results? Technical fixes like robots.txt changes take effect in days. Content and entity work usually takes months to show up consistently in AI answers.
Is AI search optimization different from SEO? Mostly no. It’s SEO with more readers to satisfy. Strong technical SEO, useful content and a trustworthy brand are what every platform rewards.
Should I block AI crawlers? Block training crawlers only if that’s your business decision. Blocking search crawlers removes you from AI answers.
Do I need llms.txt? Google Search ignores it. Lighthouse checks for it in an experimental agent audit. It’s cheap to add, but it won’t change your rankings.
Final thoughts
When this series started, AI search felt like a mystery. It isn’t. Every platform publishes its crawlers. Two of them now show you real data. And all of them reward the same things: a site machines can reach, content worth quoting, a brand they can recognize, and a reputation the rest of the web confirms.
The companies that win in AI search won’t be the ones chasing tricks. They’ll be the ones that are easiest to understand, and easiest to trust.
Thank you for following the series. If it helped, share it with someone who’s trying to make sense of AI search, and follow me for more practical technical SEO guides.
