Before LeverSolo bought its domain, we checked 10 buying-intent searches in its niche. Google showed an AI Overview on all 10. On 9 of the 10, a forum thread sat in the top two organic results.
That is the page a new site now launches into. Ahrefs compared 300,000 keywords and found that when an AI Overview appeared, the top-ranking page’s click-through rate was about 58% lower than its model predicted without one (Ahrefs, December 2025 data, published February 4, 2026). It’s a correlation from desktop data, so don’t turn it into a forecast for your site. It does tell you where attention goes: to the answer box and the handful of links inside it.
Nobody can promise you one of those links. What you can do is remove every reason an engine would skip you, write sections that are easy to quote, and measure honestly. This guide is that workflow, with a robots.txt you can copy, a before/after rewrite, a monthly check, and a checklist.
The short answer
- Google: there’s no special AI markup. Pages must be indexed and eligible to show a snippet. Normal SEO and people-first content decide the rest.
- ChatGPT, Claude, Perplexity: each runs its own search crawler. Block it and you can’t be cited there. Allow
OAI-SearchBot,Claude-SearchBot, andPerplexityBotat minimum. - Writing: open every section with its answer in two or three sentences, put comparisons in tables, and date every fact that changes.
- Identity: make it obvious who runs the site and what it covers, on the page and on an About page.
- llms.txt: optional. Google ignores it; some other tools read it.
- Measurement: once a month, check Google’s generative AI report, Bing’s AI Performance report, and a manual log of 10 to 20 real queries, each run several times.
What Google says about AI Overviews
Google’s position is short: there are no extra requirements to appear in AI Overviews or AI Mode. Its AI features documentation says a page only needs to be indexed and eligible to show in Search with a snippet, and that the same SEO fundamentals apply (checked November 9, 2026).
Its guide to optimizing for generative AI search goes further and rules out several popular tactics. Here is what it says, checked November 9, 2026:
| Popular advice | What Google’s documentation says |
|---|---|
| Add llms.txt or “AI files” | Not needed for Google Search, which ignores them. Fine to keep for other services. |
| Add special schema for AI | No special schema.org markup exists for generative AI search. Structured data is still useful for rich results. |
| Break content into small chunks | No requirement. Google can find the relevant part of a long page. There’s no ideal page length. |
| Write separate pages for AI | Pages made mainly to manipulate rankings or AI answers fall under its scaled content abuse policy. |
| Buy a tool that “uses Google’s internal signals” | No third-party tool has access to Google’s ranking or AI systems. |
Two practical points follow. First, anything that stops a snippet also stops an AI Overview link: a stray nosnippet, a tiny max-snippet, or noindex on a page you care about. Second, AI Overview traffic is already in your Search Console data. It’s counted in the Performance report under the “Web” search type, and there’s now a dedicated generative AI performance report showing impressions from AI Overviews and AI Mode, which Google says has rolled out to all websites as of August 31, 2026. It usually doesn’t appear until your site has enough impressions in those features.
So for Google, the work is ordinary SEO done well. The rest of this guide is where the non-Google engines differ.
Step 1: Let the right crawlers in
Each AI company separates the crawler that finds pages for its search answers from the crawler that collects training data. You can allow one and block the other. For citations, the search crawlers are the ones that matter.
| Company | Search crawler (allow for citations) | Training crawler | Fetches when a user asks |
|---|---|---|---|
Googlebot (AI Overviews use normal Search crawling) |
Google-Extended (a control token, not a separate crawler) |
n/a | |
| OpenAI | OAI-SearchBot |
GPTBot |
ChatGPT-User |
| Anthropic | Claude-SearchBot |
ClaudeBot |
Claude-User |
| Perplexity | PerplexityBot |
none listed | Perplexity-User |
Sources, all checked November 9, 2026: OpenAI crawler overview, Anthropic’s crawler help article, Perplexity’s bot docs, Google’s common crawlers page.
What the docs say about each choice:
- OpenAI: sites that opt out of
OAI-SearchBotwon’t be shown in ChatGPT search answers (they may still appear as plain navigational links). Each bot’s setting is independent, so you can allowOAI-SearchBotand blockGPTBot. Changes take about 24 hours to apply to search. - Anthropic: blocking
Claude-SearchBot“may reduce your site’s visibility” in user search results;ClaudeBotis the training crawler. - Perplexity:
PerplexityBotexists to surface and link sites in Perplexity’s results and isn’t used for model training.Perplexity-Usergenerally ignores robots.txt because a person asked for the page. Changes can take up to 24 hours. - Google:
Google-Extendeddoesn’t affect inclusion or ranking in Google Search. It controls whether content may be used to train future Gemini models and for grounding in Gemini Apps and Vertex AI.
A robots.txt you can copy
For a small site that wants to be cited, the simplest file is the best one. LeverSolo’s own robots.txt is this, with its own domain:
User-agent: *
Allow: /
Sitemap: https://example.com/sitemap-index.xmlIf you want to opt out of model training while staying in AI search, add named groups for the training tokens only:
User-agent: *
Allow: /
# Training opt-out. Search crawlers stay allowed by the * group above.
User-agent: GPTBot
Disallow: /
User-agent: ClaudeBot
Disallow: /
# Also limits grounding in Gemini Apps and Vertex AI, per Google's docs.
User-agent: Google-Extended
Disallow: /
Sitemap: https://example.com/sitemap-index.xmlTwo traps to avoid. A crawler follows only the most specific group that names it, so a User-agent: OAI-SearchBot group with one Disallow line replaces everything in * for that bot. And an allowlist that names five crawlers and blocks * shuts out every engine you didn’t think of.
Check your CDN too
robots.txt isn’t the only gate. Google’s AI features page lists “allowing crawling in robots.txt and by any CDN or hosting infrastructure” as a basic requirement. If your domain runs through Cloudflare, note that since July 1, 2025, Cloudflare asks every new domain at sign-up whether to allow AI crawlers (Cloudflare press release). A box ticked in a hurry on launch day can block the search crawlers your robots.txt welcomes. Open the AI Crawl Control section of your dashboard and confirm that search crawlers are allowed. Other CDNs and hosts have their own bot settings, so check yours.
Step 2: Get indexed in Google and Bing first
A page that isn’t indexed can’t be an AI Overview link, and it’s harder for other engines to find. Before any writing changes, confirm that your important pages show “URL is on Google” in Search Console’s URL Inspection tool.
Bing matters more than its search share suggests. Microsoft says Copilot may fetch information from the Bing search service to ground its answers (Microsoft Learn, checked November 9, 2026). Bing Webmaster Tools is free, and it can import your verified sites from Search Console in a few clicks.
The full setup, day by day, is in Search Console for a brand-new site: the first 30 days.
Step 3: Write sections that can be quoted alone
An AI answer quotes a passage, not a whole page. The passages that get picked up answer a question without needing the paragraphs around them. That is also what a hurried human reader wants, so this doesn’t conflict with Google’s “write for people” rule.
Four habits do most of the work:
- Answer under the heading. The first two or three sentences of each section answer the question the heading asks. Background and caveats come after.
- Put the specific thing in the answer. A price, a limit, a step count, a named option. “It depends” is never the first sentence.
- Date facts that change. “(pricing page, checked November 9, 2026)” tells the reader and any engine how fresh the number is. An undated price could be from any year.
- Use tables for comparisons and numbered steps for processes. Engines and readers can lift a row or a step without rewriting it.
Before and after
This is an illustration written for this guide, not a quote from a real site. The heading is “How much does Make cost for a solo builder?”
Before: “When it comes to pricing, there are many factors to consider. Make offers a range of plans designed to fit different needs and budgets, so it’s important to evaluate your requirements before choosing the right option for your business.”
After: “Make’s Free plan includes 1,000 credits a month. The Core plan costs $9 a month for 10,000 credits, billed annually (Make pricing page, checked November 9, 2026). If each step in a 5-step scenario uses one credit, 200 runs a month fit inside the free plan.”
The first version could sit under any pricing heading on any site. The second answers the question, names its source and date, and gives the reader a sum to check against their own use. The assumption (one credit per step) is stated, so nobody mistakes it for a quoted fact.
Don’t stop at the opener. If an engine quotes your table, the table needs its own context: a caption-like sentence above it saying what is compared and when it was checked.
Step 4: Make it clear who you are
Engines that cite sources prefer ones they can describe. A reader deciding whether to trust a quoted price wants the same thing. Make the facts about your site easy to find and consistent everywhere:
- One site name, spelled the same way on every page, in your title tags, and on your social profiles.
- An About page that says who runs the site, what it covers, and how it makes money.
- A byline and a “last updated” date on every article.
- A one-line evidence label on reviews: did you use the tool, or research it?
Structured data such as Organization and Article can describe the same facts in a machine-readable form. Google says it’s optional for AI features and must match the visible text on the page. Treat it as a description of what’s already there.
Step 5: Decide on llms.txt (optional)
llms.txt is a proposal by Jeremy Howard, first published September 3, 2024: a Markdown file at your site root with an H1 name, a short summary, and lists of links to your key pages. It isn’t a standard that the major engines have committed to. Google says plainly that Google Search ignores it and that having one will “neither harm nor help” your visibility there.
On a static site it costs about ten minutes to write and a minute to update when you publish. LeverSolo keeps one at /llms.txt for agents and tools that read it. If maintaining it ever becomes a chore, drop it before you drop anything in Steps 1 to 4.
Step 6: Run a monthly AI visibility check
AI answers change between runs, between accounts, and between locations. One screenshot proves nothing. A small, repeated log turns that noise into something you can act on.
First-party data (10 minutes)
- Google Search Console: open the generative AI performance report, if it shows for your site yet. Note impressions and the pages that earn them.
- Bing Webmaster Tools: open AI Performance (a public preview since February 10, 2026). It shows total citations across Microsoft Copilot, AI summaries in Bing, and some partner integrations, plus the grounding queries behind them. Microsoft notes the query data is a sample.
- Analytics: filter referrals from
chatgpt.com,perplexity.ai,claude.ai,copilot.microsoft.com, andgemini.google.com. Expect small numbers on a new site.
The manual log (about an hour)
Pick 10 to 20 questions a buyer would really type into a chat box, phrased the way people talk. Skip glossary questions that wouldn’t change anyone’s decision. Run each one 3 times in each engine you care about, in a fresh chat, and log every run:
date,engine,query,run,ai_answer_shown,our_url_cited,cited_domains,wrong_fact_about_us,notes
2026-12-01,google_aio,"how to get cited in chatgpt",1,yes,no,"reddit.com;ahrefs.com",,"logged out, US"
2026-12-01,chatgpt,"how to get cited in chatgpt",1,yes,no,"openai.com;example.com",,Always write the sample size next to the result. “Cited in 4 of 45 runs (15 queries x 3 runs, ChatGPT only)” is a finding. “We’re showing up in ChatGPT now” is a feeling.
What to do with the log
| What you see | What to do |
|---|---|
| An engine states something wrong about you or a tool you cover | Find the page that should state that fact and make it say so plainly in its opening lines. Fix this first. |
| You’re absent, and a weaker page is cited | Check that your page is indexed and that its section opener answers the query directly. |
| A forum or review site is cited for every run | That source is part of the answer. Consider whether a useful, disclosed contribution there makes sense. |
| You’re cited in 1 or 2 runs out of 45 | Treat it as noise. Compare the next three months before changing anything. |
Keep the same query list for at least three months so the months can be compared. Add new queries in a separate block.
The checklist
Copy this into your publishing checklist:
AI CITATION CHECKLIST (per site, once)
[ ] robots.txt allows OAI-SearchBot, Claude-SearchBot, PerplexityBot (a plain "User-agent: * Allow: /" does)
[ ] CDN or host bot settings don't block those crawlers
[ ] Sitemap listed in robots.txt and submitted in Search Console and Bing Webmaster Tools
[ ] About page states who runs the site, what it covers, how it earns
[ ] llms.txt: decided (optional)
AI CITATION CHECKLIST (per article)
[ ] Indexed: URL Inspection shows "URL is on Google"
[ ] No nosnippet / max-snippet / noindex on the page by accident
[ ] Each section's first 2-3 sentences answer its heading
[ ] Every price, limit, and date-sensitive fact names its source and check date
[ ] Comparisons in tables, processes in numbered steps
[ ] Byline and "last updated" date visible
[ ] Any structured data matches the visible text
MONTHLY
[ ] Search Console generative AI report noted
[ ] Bing AI Performance noted
[ ] AI referrals in analytics noted
[ ] Manual log: same 10-20 queries x 3 runs per engine, sample size written down
[ ] Wrong facts about you fixed on the page that should state themWho this is for
It fits you if you run a small content or affiliate site, your pages are already indexed, and you want a routine that takes about an hour and a half a month.
It doesn’t fit you if you’re looking for a trick that gets a new page cited before it ranks or is even indexed. Every engine in this guide first has to find and trust the page, and that part is ordinary search work. Start with the content system that makes pages worth citing.
Quick answers
Do I need special schema or an llms.txt file to appear in AI Overviews? No. Google’s documentation says there are no extra requirements and that Google Search ignores llms.txt. The page has to be indexed and eligible for a snippet.
If I block GPTBot, will I disappear from ChatGPT? Not from search answers. OpenAI documents GPTBot (training) and OAI-SearchBot (search) as independent settings, so you can block the first and allow the second.
How do I know if ChatGPT cites my site? OpenAI’s crawler docs describe no publisher report like Search Console. Use your analytics referrals from chatgpt.com plus a manual log of real queries run several times. Bing’s AI Performance report covers Copilot, and Search Console’s generative AI report covers AI Overviews and AI Mode.
More in AI for Content & SEO. Next, set up the measurement side with Search Console for a brand-new site, or build the articles worth citing with the AI content system for solo sites.