AI Search Optimization for Businesses: The Complete Guide to Getting Recommended by ChatGPT, Gemini, and Google AI

Take the question a customer would actually ask - “What’s a good accountant for a small business?” - paste it into an AI assistant, and see whether your business is in the answer. That’s the entire question this guide is about. Everything else - AEO, GEO, llms.txt, schema, citation rate - is a way of getting closer to a yes.

The reason this has become a board-level and kitchen-table topic at the same time is that the way people look things up genuinely changed. In May 2026, Google reported that AI Mode had passed a billion monthly active users globally, that AI Mode queries had “more than doubled every quarter since launch,” and that the average AI Mode search is three times the length of a traditional Google query. Those are Google’s own US-focused figures, published 19 May 2026 - five months before this article was written, so treat them as a snapshot rather than a live count. But the direction isn’t in doubt, and it has a practical consequence: longer, conversational questions are the queries that trigger AI answers, and they aren’t the queries most businesses were built to rank for.

Here is the honest version of what works, what doesn’t, and when the whole exercise isn’t worth your money.

What “being recommended” actually means

Start here, because most bad AI search advice begins by describing a ranking system that doesn’t exist.

When you ask ChatGPT, Gemini, Copilot or Perplexity a question, the system doesn’t “rank your website.” It does something closer to this: it works out what information it needs, goes and retrieves candidate sources, reads them, checks those sources against each other, and then writes an answer that cites them. Google’s own description of this in its current optimization guide calls it retrieval-augmented generation, or grounding - “relying on our core Search ranking systems to retrieve relevant, up-to-date web pages from our Search index,” after which its systems “review the specific information from those retrieved pages to generate a more reliable and helpful response.”

So the unit of success isn’t a position. It’s a citation: your page or your business record being named, linked, or quoted inside an answer the model wrote and the model didn’t entirely write by itself. Microsoft’s guide makes the same point more bluntly in a different direction - “there’s no secret sauce that guarantees selection in AI answers.” OpenAI’s help documentation puts it plainly: “Placement is not guaranteed.”

You can’t choose which of ten sources an assistant cites. You can make sure your business is in the running. You can also write claims that still make sense when they’re lifted out of the page.

One more mechanism matters, because it explains why a single perfect page often isn’t enough. Both Google and Microsoft describe a step where the original question is split into several related sub-queries. Google’s published example: if someone asks “how to fix a lawn that’s full of weeds,” fan-out queries might include “best herbicides for lawns,” “remove weeds without chemicals,” and “how to prevent weeds in lawn.” In that worked example, a site that addresses only the first phrasing is competing for one of three sub-chances rather than all of them. This is also why Google’s guide warns that creating a separate page for every fan-out query, primarily to manipulate rankings, “violates Google’s scaled content abuse spam policy.”

Five surfaces, one website

For a business owner, the most useful reframe is that this isn’t five search engines. It’s one website being read by five front doors with different access rules.

Google AI Overviews and AI Mode are built on the same index as Google Search. Eligibility is the baseline: “a page must be indexed and eligible to be shown in Google Search with a snippet.” Google also says AI Overviews “are only shown when our systems determine that it is additive to classic Search, and as such, often don’t trigger” - so absence from AI Overviews is normal and isn’t, by itself, a problem with your site.

Gemini sits inside Google’s surfaces rather than beside them. If you sell physical products, Google’s Merchant Center documentation is unusually direct about this: eligible free listings mean products “can appear at no cost on Google Search, Google Maps, Gemini, YouTube, the Shopping tab, Google Images, and Google Lens.”

ChatGPT search is a separate crawler story. OpenAI runs several user-agents, and only one of them decides search visibility. This is the single most consequential technical detail for newcomers, because the intuitive move - block GPTBot - is the wrong one. OpenAI’s crawler documentation is explicit: “Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links.” GPTBot is the training crawler, and each setting is independent. Changes take effect quickly - “it can take ~24 hours from a site’s robots.txt update for our systems to adjust.”

Perplexity splits the same way. PerplexityBot “is designed to surface and link websites in search results on Perplexity. It is not used to crawl content for AI foundation models.” Perplexity-User is the user-triggered fetcher, and Perplexity is candid that “since a user requested the fetch, this fetcher generally ignores robots.txt rules” - which means a firewall rule is the only real control there.

Claude uses three named robots: ClaudeBot for model development, Claude-User for user-initiated requests, and Claude-SearchBot for search indexing. Anthropic states that disabling Claude-SearchBot “prevents our system from indexing your content for search optimization, which may reduce your site’s visibility and accuracy in user search results.”

Microsoft Copilot is the odd one out, because it isn’t a separate web index. It’s built on Bing’s, which means Copilot visibility is largely a function of ordinary Bing indexing - and, per Microsoft, of the answer-generation layer on top of it. Its AI Performance documentation is the most explicit of any of these platforms about what a citation is and isn’t, which we’ll come back to in the measurement section. For local businesses, Microsoft’s AI Performance post points to Bing Places for Business as the equivalent of a Business Profile, so that “key details such as address, hours, and contact information remain current and eligible for inclusion in AI-generated responses.”

Engine	Crawler that controls search inclusion	Crawler that controls training	Change propagation
Google	Googlebot (same as Search)	Google-Extended (separate opt-out)	Days to months for recrawl
ChatGPT	OAI-SearchBot	GPTBot	~24 hours
Perplexity	PerplexityBot	PerplexityBot is not used for training	Up to 24 hours
Claude	Claude-SearchBot	ClaudeBot	Not stated publicly
Copilot	Bingbot (same as Bing Search)	Not separately documented here	Sitemap revisit “at least once per day”

The practical takeaway is blunt. Check your robots.txt and your firewall before you check anything else. If you blocked one of these crawlers during a 2023 panic about AI scraping, you may have quietly removed yourself from an entire channel. Note that Anthropic warns that blocking by IP address “may not work correctly or persistently guarantee an opt-out,” and that Perplexity’s own documentation asks you to whitelist its bots explicitly if you run Cloudflare or AWS WAF. A robots.txt entry you never wrote is doing more damage than any content gap.

There’s also a newer layer. Google’s guide points at “browser agents” that “may access your website to gather the data they need,” including by “analyzing visual renderings (like screenshots), inspecting the DOM structure, and interpreting the accessibility tree,” and notes that protocols like Universal Commerce Protocol are emerging. This is early. Treat it as a reason to keep your site structurally honest, not as a project.

The AI Readiness Baseline

Everything above is context. This is the part you can actually use.

It has twelve checks across four dimensions, each scored 0 (broken or missing), 1 (present but weak), or 2 (solid). You should be able to finish it in about fifteen minutes. Be honest rather than generous; a flattering score here wastes your budget.

Dimension A - Can the engines reach you?

1. Your robots.txt doesn’t block any search crawler. Check it against the table above. Most sites pass this by accident, not by decision.
2. Your CDN or WAF isn’t silently blocking AI crawlers. Check access logs for OAI-SearchBot, PerplexityBot, and Claude-SearchBot. If you have never seen one, you can’t assume one was allowed.
3. You have an XML sitemap, it’s referenced in robots.txt, and its dates are honest. Google’s own guidance for Bing’s crawler is blunt: “Avoid setting lastmod to the time your sitemap was generated unless the content on that URL was actually updated.”

Dimension B - Is there anything worth quoting?

4. You have at least one page containing something only you could write - your own data, your own process, your own failures, your own photographs. Google’s test is the sharpest available: “a first-hand review provides a unique perspective based on personal experience, whereas a summary of existing content simply restates information already available elsewhere.”
5. Your key pages state specifics, not adjectives. Microsoft’s guidance is concrete: say “42 dB dishwasher designed for open-concept kitchens,” not “quiet dishwasher.” Vague terms like “innovative” or “eco” carry no extractable meaning.
6. Your people are attributable. Does a reader know who wrote this and what they do? Google asks whether content “present[s] information in a way that makes you want to trust it,” and warns pointedly that “Fabricating creator profiles (such as AI-generated headshots, made-up names, or false credentials) … is a form of deception.”

Dimension C - Does the record match everywhere?

7. Your business name, address, hours and phone number are identical across your website, your Business Profile, your Merchant Center account and your directory listings. Google’s Business Profile guidance is unusually literal about this - no marketing taglines in the name field, no location qualifiers, no store codes. Microsoft points local businesses at Bing Places for Business for exactly the same reason: keeping “address, hours, and contact information … current and eligible for inclusion in AI-generated responses.”
8. If you sell products, your Merchant Center feed is complete and your free listings are on. Google notes free listings are on by default in most cases, but the required attributes - id, title, link, image link, price, description, availability - are strict, and shipping configuration is mandatory in a long list of countries including the US, UK, Canada, India and Australia.
9. Your structured data matches what is visibly on the page. Google’s guidance is that structured data “isn’t required for generative AI search,” but if you do use it, “Making sure your structured data matches the visible text on the page” remains standard practice.

Dimension D - Could you prove it worked?

10. You’re verified in Search Console and can find the Generative AI performance report. Google rolled it out to all websites worldwide on 31 August 2026, and it covers AI Overviews and AI Mode. If it’s missing, Google’s own explanation is unfashionably simple: “your site hasn’t received enough impressions in generative AI features.”
11. You’re verified in Bing Webmaster Tools and can find AI Performance. This shows Total Citations, Average Cited Pages, grounding queries, page-level citation activity and trends.
12. You have a recurring thirty-minute review of what AI assistants actually say about you. Not a tool watching other people run your prompts for you - someone on your team asking your real customer questions, monthly, and reading the answers.

Reading your score

0–7 - Not in the game. Something in Dimension A is broken and nothing else matters yet. Fix crawling and firewall rules before spending a single content hour. This is usually a one-afternoon job.

8–14 - Present but unproven. You’re online and probably invisible. Focus entirely on Dimensions B and C. This is where most businesses should be, and it’s where the highest-return work lives.

15–19 - Competitive basics are in place. Stop adding surface area and start improving depth. The next gain is in attribution and in third-party corroboration, not in more pages.

20–24 - Extend, don’t re-platform. At this point your marginal return is in staying current on platform changes, in freshness, and in the emerging agentic surfaces - not in rebuilding anything.

This score measures readiness. A high score means you’re eligible, not that you’ll be cited. The weighting favours checks you can make cheaply. If Dimension A is weak, more content won’t fix it. Anyone selling you a content strategy for that problem is starting in the wrong place.

Why you cannot simply review yourself into the answer

This is the section that most surprises business owners, and it’s the one I would put at the centre of any first attempt at AI search work.

The instinct is obvious: put your best testimonials, your star rating and your customer quotes on your own site, mark them up, and let the assistants find them. Google’s structured data policy shuts this down. For local businesses and organizations, Google states: “If the entity that’s being reviewed controls the reviews about itself, their pages that use LocalBusiness or any other type of Organization structured data are ineligible for star review feature.” The example given is pointed - a review about entity A placed on entity A’s own site, whether marked up directly or pulled in through a third-party widget, doesn’t qualify.

The rest of the policy is equally firm: “Ratings must be sourced directly from users.” “Don’t rely on human editors to create, curate, or compile ratings information for local businesses.” And “Don’t aggregate reviews or ratings from other websites.”

AI visibility is structurally biased toward businesses that already exist in other people’s records. A business with a poor website and strong third-party presence is in a better position to be cited by an assistant than a business with an excellent website and no external footprint. That isn’t an argument for ignoring your site. It’s an argument for understanding that the cheapest, highest-return AI search work for most businesses isn’t publishing - it’s making sure the version of you that already exists on the internet is accurate, consistent and current.

Before you write anything new, check your existing directory listings, industry profiles, review platforms, supplier pages, and association memberships. Do they agree on your name, address, and services? Then leave fake corroboration alone. Google’s policy rules out fake reviews and reviews given for an undisclosed benefit. The damage to your reputation is more serious than a lost citation.

The technical layer, in plain language

You don’t need to become a developer to pass this section. You do need to know what you’re asking your web team for.

Making sure content is findable and readable. Google restates the basics for AI surfaces as: crawling is allowed in robots.txt and by any CDN or hosting infrastructure; content is easily findable through internal links; important content is available in textual form. If your core information lives inside an accordion, a tab, or an image, it isn’t reliably available. Microsoft’s guidance says the same thing from the other direction - “AI systems may not render hidden content, so key details can be skipped” - and adds that relying on PDFs for core information is a mistake because they “often lack the structured signals … that HTML provides.”

Structured data, correctly understood. Google’s current position is unusually clear, and it’s worth quoting in full because it directly contradicts most of what is sold under this heading: “Structured data isn’t required for generative AI search, and there’s no special schema.org markup you need to add.” What it is good for is disambiguating you. Google’s Organization documentation recommends adding properties like name, logo, url, sameAs, contactPoint and address, and notes that using “the same name and alternateName that you’re using for your site name” helps. The sameAs field is the interesting one for this topic: it’s where you point at the other authoritative records of yourself - your Business Profile, your directory listings, your social profiles. It’s a machine-readable assertion that those records describe the same entity, which is exactly the corroboration problem described above.

Telling the engines you changed something. IndexNow is the most useful piece of new plumbing here. You create a key file on your host, then notify participating engines - including Bing - when a URL is added, updated or removed. Batch submissions take up to 10,000 URLs per POST. Bing documents that sitemaps are re-checked “at least once per day” and that IndexNow is how you do better than that. Microsoft’s AI Performance announcement is explicit about why it matters for AI specifically: faster discovery “helps ensure that AI systems reference the most current version of a page when generating answers.”

The llms.txt question, answered without hype. This is where the industry is most confidently wrong, so it’s worth being precise. llms.txt is a real, documented proposal - llmstxt.org now describes version 2 of the spec, and lists platforms that generate the file automatically, including Mintlify, GitBook, Yoast SEO and Wix. The proposal’s own page claims “thousands of sites publish an llms.txt file.”

Against that: Google’s optimization guide, last updated 10 July 2026, mythbusts it directly - “You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn’t use them.”

Chrome Lighthouse does ship an llms.txt audit. Chrome’s own documentation, last updated 5 May 2026, describes it as flagging pages “if a server error occurs when attempting to retrieve the llms.txt file,” and - this is the part that matters - “If the file is not provided by the server (resulting in a 404), the audit is marked as Not Applicable (N/A), as providing the file is optional at the moment.”

One Google team says you don’t need the file. Another ships a check that passes even when the file is missing. I wouldn’t build an llms.txt file to chase AI visibility. If you run documentation, it may help coding agents at little cost. That’s a reason to improve developer experience. The Lighthouse check doesn’t turn it into a visibility trick, and it can’t fail you for leaving the file out.

Measurement: what you can see and what you cannot

Measurement has improved faster than the advice about it, and the gap is where most businesses get misled.

The genuinely new reporting is in Search Console. Google’s Generative AI performance report became available to all websites worldwide on 31 August 2026. It covers AI Overviews and AI Mode, and you can slice it by Pages, Countries, Dates and Devices, filtering by search type. Two honest caveats: the report counts impressions, which Google defines as how many times links to your site were shown in a generative feature - not clicks, and not revenue. And the usual Search Console row limit applies. Google’s note that if the report is missing, the site simply hasn’t accumulated enough AI impressions, is worth reading twice, because it means absence of data is the expected state for most businesses, not a sign of failure.

There’s a reconciliation worth making, because two Google pages appear to disagree. The 2025 guidance page says there are “no additional technical requirements” for AI features. The current guide adds a condition: “a site must be included in Search generative AI features in Search Console to be eligible for display in generative AI features.” The current guide is the one to follow, and it’s also the newer one - its last-updated date is 10 July 2026. Note also that the older page remains online and still says AI feature traffic is reported inside the ordinary Performance report under the “Web” search type; that’s now incomplete, since a dedicated report exists.

Bing Webmaster Tools is the second real reporting surface. AI Performance, in public preview since 10 February 2026, shows Total Citations, Average Cited Pages, grounding queries, page-level citation activity and visibility trends - specifically across Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations. Microsoft is refreshingly blunt about what these numbers mean: Average Cited Pages “does not indicate ranking, authority, or the role of any page within an individual answer,” and page-level data “reflects how often pages are cited, not page importance, ranking, or placement.”

In October 2026 Microsoft added a beta feature called Topics Portfolio, which groups your grounding queries into thematic clusters and shows, for each, Topic Size, Sites Cited and Queries - plus “Opportunity Queries” where other sites appear and yours doesn’t. It comes with honest warnings attached: topic classification is AI-generated and imperfect, “Topics with insufficient site presence or insufficient ecosystem activity are not shown, so smaller sites may see a shorter list or may not even see this feature,” and Microsoft asks you to “Treat the numbers as directional signals for planning, not as a precise ledger.” Small businesses should expect this tab to be empty or nearly empty. That’s a size threshold, not a quality verdict.

Which means the honest measurement stack for most businesses is: Search Console’s Generative AI report, Bing’s AI Performance, your existing web analytics, and a monthly manual audit of what assistants actually say. That last one is unglamorous and, in my view, the most informative of the four, because it measures the thing you actually care about.

On traffic, the honest answer is that the data is old, mixed, and contested. The most-cited independent study is Pew Research Center’s analysis of browsing data from 900 US adults across 68,879 Google searches in March 2025, published as a short read on 22 July 2025 - about fifteen months before this article. Its findings: users who saw an AI summary clicked a traditional result 8% of the time, versus 15% for those who didn’t; they clicked a link inside the summary just 1% of the time; and they ended their session entirely 26% of the time versus 16%. Wikipedia, YouTube and Reddit together accounted for 15% of the sources cited in those summaries.

Two things follow. First, that’s a fifteen-month-old snapshot of a fast-moving surface, and I wouldn’t present it as a current forecast. Second, it cuts against the simpler doom narrative, and even Google’s own framing does the same: on the AI features page, Google says “when people click from search results pages with AI Overviews, these clicks are higher quality (meaning, users are more likely to spend more time on the site).” Fewer clicks, better clicks is a plausible reading of both, but nobody has established which applies to your business.

Note too what Pew found about which queries trigger summaries: 8% of one- or two-word searches produced one, rising to 53% for searches of ten words or more. Short navigational searches largely still behave like search. It’s the long, conversational, comparison-shaped question that has changed.

A realistic timeline

The honest answer is that most of the work isn’t slow, and the part that’s slow isn’t the part agencies sell.

Week one. Audit robots.txt, CDN and firewall rules. Fix what is blocking. This is days rather than months, and it’s close to the only step with an unambiguous pay-off.

Weeks one to four. Reconcile your business records: Business Profile, Merchant Center, Bing Places, and the third-party listings that describe you. Fill gaps in Organization markup, especially sameAs. Complete your Merchant Center feed if you sell products.

Weeks two to twelve, ongoing. Publish a small number of genuinely first-hand pieces. Not thirty posts. Three or four that contain something no competitor can copy. Google’s guidance on volume is worth internalising: “a high quantity of pages doesn’t make a website higher quality or more relevant to users,” and if you’re adding or removing content “primarily because you believe it will help your search rankings overall by somehow making your site seem ‘fresh’ … (No, it won’t).”

Months three to twelve. Third-party corroboration compounds slowly. Industry directories, association listings, supplier pages, genuine editorial coverage. This is where a business that already has a real market presence pulls away from one that doesn’t.

Six to twelve months. Any visible AI citation trend. Google’s own guidance on how long recrawling takes after a control change is “anywhere from several days to several months, depending on how often our systems determine a page needs to be refreshed.” Set expectations accordingly, and set them with your team, not with your vendor.

Do it in-house or hire it out?

Gate 1 - the evidence gate. Do you have six months or more of clean first-party data, and does somebody read it monthly? If you can’t answer “what did AI Overviews do to our traffic last month” with a real number, you aren’t ready to buy a strategy. You need to be able to tell whether one worked.

Gate 2 - the ownership gate. Is there one named person accountable for this for the next two quarters? Not a committee, not an agency relationship without an internal owner. Evidence for why: every platform here is changing on a schedule you don’t control, and unowned work is unmaintained work within about four months.

Gate 3 - the production gate. Can you publish at least one substantial, genuinely first-hand piece a month without a new hire? If the honest answer is no, you don’t have a distribution problem yet - you have a production capacity problem, and no amount of optimisation fixes that.

The rule: fail two of the three gates, and don’t commission a bespoke programme. Instead, buy tools or a fixed-scope engagement that ends in a documented internal capability, and spend the difference on the parts that always work. In practice, for most businesses at gate-failure, that means fixing access and records first - those are days of work - and only then writing.

Three conditions where hiring is clearly the wrong call. Nobody owns it. There’s no measurement, so nobody can tell whether it worked. Or the site’s fundamentals are broken - a site that doesn’t load, isn’t crawlable, or has no products or services described anywhere - in which case an AI search strategy is treating a symptom.

Two conditions where hiring is clearly right. You’re in a category where third-party evidence is the entire game - regulated, medical, legal, financial - because you can’t generate your own authority and must borrow it. Or you have a genuine multi-surface operation across several markets and languages, where the coordination overhead exceeds what a small in-house team can absorb.

On budget

Be careful with numbers in a vendor’s pitch. Almost every figure circulating in this category comes from someone selling the service it describes. One Microsoft Advertising post makes an AI referral growth claim using two numbered sources you can’t see on the page. I’d ask for those sources before using the number.

What can be said honestly is that the floor is very low and the ceiling is unbounded.

The floor: Dimension A of the Readiness Baseline is roughly one afternoon for someone comfortable with a robots.txt file and server logs. Dimension C is a few days of spreadsheet work. Neither requires a vendor.

The middle: structured data, Merchant Center feeds, sitemaps and IndexNow integration are days of competent technical work, not projects.

Original reporting, first-hand testing, and expert authorship cost money. They did before AI search, too. If your plan relies on publishing more instead of saying something useful, it will fail. Blaming AI search won’t fix the plan.

Measure effort in hours and attribute the AI share honestly. If your total programme is, say, two hundred hours over six months, be curious about any proposal that costs a comparable number of hours while producing no published work.

The mistakes that cost businesses the most

Building an llms.txt file because a tool told you to. Google says it doesn’t use the file. Chrome’s Lighthouse audit marks it Not Applicable when absent. If you built one, it isn’t hurting you - but it isn’t helping you either, and it may have displaced work that would have.

Publishing one page per AI-generated variant of the same question. This is explicitly named as scaled content abuse in Google’s current guide.

Trying to self-certify your reputation. Self-serving reviews are ineligible for Google’s star review feature, ratings must come directly from users, and Google’s policy explicitly rules out fake or undisclosed incentivised reviews.

Blocking AI crawlers out of caution. Blocking GPTBot while assuming you’re still in ChatGPT search is a common and expensive misunderstanding. Search inclusion is governed by a different crawler.

Writing vague marketing copy and calling it content. If a sentence has no fact in it, an assistant has nothing to lift. “Innovative,” “leading,” and “next-generation” aren’t extractable.

Hiding your answers in tabs, accordions or PDFs. Documented as a real parsing failure mode by Microsoft’s guidance, and Google’s baseline requirement is textual availability.

Chasing AI traffic and forgetting that Google Search still exists. Google is explicit that plenty of content “thrives in Google Search (including generative AI experiences) without any overt SEO at all,” and that generative AI features are “rooted in our core Search ranking and quality systems.” This isn’t a replacement for SEO. It’s SEO, plus two new front doors.

Buying a guarantee. No vendor can promise citation placement, and the platforms themselves disclaim it. OpenAI: “Placement is not guaranteed.” Microsoft: “nothing guarantees selection.” A contract that promises otherwise is a contract that will be disputed.

Where the documentation genuinely disagrees

Does content need to be restructured into modular pieces? Microsoft’s guidance says assistants “break content down … into smaller, structured pieces” and gives detailed advice on headings, Q&A blocks, lists and tables. Google’s current guide says the opposite: “‘Chunking’ content: There’s no requirement to break your content into tiny pieces for AI to better understand it.” Both can be true - structure that helps a human reader is good practice regardless - but they aren’t the same claim, and you should notice that the vendor with a marketing product in this space is the one pushing modularity.

Is there an additional technical requirement for AI inclusion? Google’s 2025 page says no. Google’s 2026 page adds a Search Console eligibility condition. Cite the current one.

Do AI clicks hurt or help? Pew’s 2025 browsing data points one way; Google’s own statement about “higher quality” clicks points another. Unresolved, and probably site-dependent.

Does llms.txt matter? One Google team says no, Chrome ships an audit for it. Resolved in practice - it’s optional - but the disagreement is real.

Your first ninety days

If you read nothing else, do this.

Week one: read your robots.txt and your server logs. Confirm OAI-SearchBot, PerplexityBot and Claude-SearchBot are neither blocked nor blocked by accident. Fix it.

Weeks one to four: complete the Business Profile, get a Bing Places listing, check whether your Merchant Center free listings are on and your feed is complete. Add Organization markup with sameAs links. Reconcile your name, address, hours and phone across every third-party listing you can find.

Weeks two onward: publish three things that contain something only you know. Then keep publishing them monthly, with a named author and a clear reason the page exists.

Ongoing: open Search Console’s Generative AI performance report and Bing’s AI Performance once a month. Run your ten customer questions through ChatGPT, Gemini, Copilot and Perplexity and save the answers. Watch what changes.

At day ninety, run the Readiness Baseline again. Compare the score with your manual audit. If it has improved and your business appears where it didn’t a quarter ago, keep going. If the score hasn’t moved, I’d look at Dimension A or Dimension C before commissioning more content. More writing won’t diagnose those problems.

Sources and References

• Google Search Central, Google’s Guide to Optimizing for Generative AI Features on Google Search - last updated 2026-07-10
• Google Search Central, AI Features and Your Website - last updated 2025-12-10
• Google Search Central, Creating Helpful, Reliable, People-First Content - last updated 2026-10-05
• Google Search Central, Google Search’s Guidance on Generative AI Content - last updated 2026-10-01
• Google Search Central, Organization (Organization) structured data - last updated 2026-09-08
• Google Search Central, Review Snippet (Review, AggregateRating) Structured Data - last updated 2026-09-08
• Google Search Central, Intro to Product Structured Data on Google - last updated 2025-12-10
• Google Search Console Help, Generative AI performance report (Search) - worldwide rollout 2026-08-31
• Google Business Profile Help, Guidelines for representing your business on Google
• Google Merchant Center Help, Free listings for products
• Google, How AI Mode is changing the way people search in the U.S. - 2026-05-19
• OpenAI, Overview of OpenAI Crawlers
• OpenAI Help Center, Searching the web with ChatGPT
• Anthropic Support, Does Anthropic crawl data from the web, and how can site owners block the crawler? - 2026-04-07
• Perplexity, Perplexity Crawlers
• Bing Webmaster Blog, Introducing AI Performance in Bing Webmaster Tools Public Preview - 2026-02-10
• Bing Webmaster Blog, Introducing Topics Portfolio (Beta) in AI Performance - 2026-10-13
• Bing Webmaster Blog, Keeping Content Discoverable with Sitemaps in AI Powered Search - 2025-07-31
• Microsoft Advertising, Optimizing Your Content for Inclusion in AI Search Answers - 2025-10-08
• IndexNow, Documentation
• llmstxt.org, The /llms.txt file, v2
• Chrome for Developers, llms.txt - Lighthouse - last updated 2026-05-05
• Pew Research Center, Do people click on links in Google AI summaries? - 2025-07-22, based on March 2025 browsing data