How to Show Up in AI Search

Why AI search visibility matters now

AI answers are no longer a niche surface — they're where a huge share of buyers now start. Google confirmed at I/O that AI Overviews reach 2.5 billion users a month, and tracking data shows they now trigger on roughly half of searches. ChatGPT, meanwhile, sits somewhere around 800 million to 900 million weekly users depending on the month. When someone asks one of these engines a question in your category, it doesn't hand back ten links to pick from. It writes one answer and names a few brands. You're either in that answer or you're not.

The traffic that does come through converts. Similarweb clickstream data puts ChatGPT referral traffic at roughly a 7.1% conversion rate — second only to paid search and ahead of organic, direct, social and email (treat the figure as a clickstream estimate). It's lower volume than classic search but far higher intent: people arriving from an AI answer have already been pre-sold by the engine. Gartner has projected traditional search volume could fall by around a quarter as users shift to answer engines, which means the question isn't whether AI search matters yet — it's whether you're visible before your competitors lock in the citations.

The three gates: crawlable, recognized, extractable

Most advice on AI search is a grab-bag of tips with no structure. Here's the structure. To appear in an AI answer your brand has to pass three gates, in order — each one a prerequisite for the next:

1. 1**Crawlable** — the engine can fetch your pages. If your robots.txt blocks its crawler, or you're absent from the index it draws on, nothing else matters. You can't be cited if you can't be read.
2. 2**Recognized** — the engine knows who you are and trusts you. Models weight brand recognition and third-party authority heavily. An unknown brand with perfect on-page content still loses to a known one.
3. 3**Extractable** — your content can be lifted as a clean, self-contained answer. Even a trusted, crawlable page won't get quoted if the model has to guess at your meaning. Structure and a direct answer block decide this.

Think of it as a filter. Gate one is binary plumbing. Gate two is reputation. Gate three is craft. The rest of this guide takes each gate in turn, with the evidence and the exact tactics — then shows you how to measure whether it's working.

Gate 1 — Crawlable: let the AI bots in

AI engines can't cite what they can't fetch. This is the gate people quietly fail most often, usually because a security plugin or a copy-pasted robots.txt is blocking the very crawlers they want. Different bots do different jobs, and blocking the wrong one silently removes you from AI answers.

Know which bot does what

The checklist for gate 1

• Open your robots.txt and confirm you're **allowing** GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Googlebot/Google-Extended and BingBot. See our AI crawlers reference for the exact directives.
• Verify your site in **Bing Webmaster Tools** and **Google Search Console** — being in both indexes is the foundation for ChatGPT and Gemini/AI Overviews respectively.
• For Google's AI features specifically, a page must be **indexed and eligible to show with a snippet** — that's the standard Search technical requirement, nothing exotic.
• Remember robots.txt changes take **about 24 hours** to register with OpenAI's crawlers. Change it, then re-check tomorrow, not in five minutes.
• Don't over-rely on llms.txt here — it's a useful docs-discovery convenience, not a crawl-permission or ranking mechanism (more on that in the mistakes section).

Gate 2 — Recognized: become a trusted entity

Crawl access gets you in the door. Recognition gets you named. AI engines apply a quality and trust filter that closely parallels Google's E-E-A-T (experience, expertise, authoritativeness, trustworthiness). Reverse-engineering studies of Perplexity, for instance, suggest it runs an entity-clarity and authoritativeness gate before a source is eligible to be cited. The model has to be confident about who you are and that you're worth recommending.

Earned beats owned

Here's the finding that reorganizes most content strategies: AI engines favor authoritative third-party sources over brand-owned content. Independent studies of where AI citations actually come from — Semrush's three-month analysis, Profound, and Peec AI's look at 30 million sources — keep landing on the same short list of most-cited domains: Reddit, YouTube, LinkedIn, Wikipedia and Google's own properties. Across these studies the top five domains account for roughly 38% of all AI citations.

LinkedIn is the standout mover. Profound found LinkedIn roughly doubled its citation frequency and is the single most-cited domain for professional queries across all six major AI platforms. The lesson isn't "spam LinkedIn" — it's that your presence in the places engines already trust does more for your AI visibility than another page on your own blog.

The recognition checklist

• **Fix your entity signals.** Keep your NAP (name, address, phone) consistent across Google Business Profile, Bing Places, Apple Maps, Yelp and Facebook. Inconsistency makes the engine unsure you're one entity, and uncertainty fails the gate.
• **Build earned media.** Get mentioned, reviewed, interviewed and linked by sources the engines already cite. Third-party coverage is worth more than self-published claims.
• **Show up where engines look.** A credible presence on Reddit, YouTube, LinkedIn and Wikipedia-eligible coverage feeds directly into the most-cited domain pool.
• **Demonstrate real E-E-A-T.** Named authors with credentials, clear publish/update dates, and genuine first-hand expertise on the page — the same signals Google rewards, the engines reward too.
• **Publish original research or proprietary data.** Nothing earns a citation like being the source everyone else has to cite. More on this in the next section.

The brands that win AI citations aren't the ones with the most pages — they're the ones the rest of the web already talks about. Earned authority is the moat.

Gate 3 — Extractable: make content easy to lift

You can be crawlable and trusted and still not get quoted — because the model couldn't cleanly extract an answer from your page. This gate is about craft, and it's the one with the strongest hard evidence behind it.

The GEO research: what actually moves the needle

The most rigorous evidence we have is the peer-reviewed GEO paper (Aggarwal et al., KDD 2024). The authors built GEO-bench — 10,000 queries across 25 domains — and measured how different content techniques changed a page's visibility inside generated answers. The headline: well-chosen tactics boosted visibility by up to roughly 40%. Crucially, they also found what hurts.

Read that bottom row twice. Keyword stuffing — the reflex of a decade of SEO — measurably *lowered* visibility in generative engines. The engines reward content that reads like a credible human wrote it for other humans, and they punish the opposite.

Front-load a direct answer

On every priority page, open with a self-contained answer of about 40–60 words that a model could lift verbatim and be correct. Vendors report front-loading like this can lift citation likelihood substantially (treat the exact percentages as directional). The logic is simple: you're handing the engine a ready-made, accurate quote so it doesn't have to assemble one — and risk paraphrasing you wrong.

Structure for machines and people

• Use **clean semantic HTML** — real headings, lists and tables — so the model parses your meaning instead of guessing at it.
• Keep content **fresh**. Perplexity in particular favors material from roughly the last 6–18 months on time-sensitive topics. Update and re-date your cornerstone pages.
• Add **schema** (Article, FAQ) for rich-result eligibility — but treat it as supporting, not mandatory. Google is explicit that schema is **not required** for AI features.
• Do **not** chop articles into tiny "AI-digestible" fragments. Google says its systems understand multiple topics on one page; fragmenting hurts the reader without helping the engine.

The step-by-step playbook

Put the three gates together and you get a concrete sequence. Work it top to bottom — early steps unblock the later ones.

1. 1

   Open the gates

   Allow the AI crawlers in robots.txt, then verify your site in Bing Webmaster Tools and Google Search Console. Re-check robots.txt ~24h later, since OpenAI's crawlers take a day to register changes.

2. 2

   Front-load answers

   Add a self-contained 40–60 word answer block at the top of every priority page — question-shaped heading, complete answer, no teasing.

3. 3

   Add the three high-ROI levers

   On those pages, cite credible sources, add real statistics, and include expert quotations. These were the top performers in the GEO study — combine all three.

4. 4

   Publish original data

   Run a survey, share your benchmarks, release proprietary numbers. Original research gives engines a reason to cite you over the lookalikes paraphrasing each other.

5. 5

   Keep it fresh

   Update and re-date cornerstone content on a cadence. Time-sensitive topics decay fastest — Perplexity leans on the last 6–18 months.

6. 6

   Build earned presence

   Earn mentions and coverage on Reddit, LinkedIn, YouTube and Wikipedia-eligible sources — the domains the studies show AI engines cite most.

7. 7

   Add supporting schema

   Mark up Article and FAQ content for rich results. Helpful for classic search; supporting (not required) for AI features.

8. 8

   Measure and iterate

   Track mentions and share of voice per platform on a schedule, change one thing, re-measure. AI answers drift, so this is a loop, not a launch.

Common mistakes that keep you invisible

AI search is new enough that a lot of confident advice is wrong, and some of it actively hurts. The ones worth unlearning:

• **Thinking llms.txt is a ranking signal.** Google is explicit: "You don't need to create new machine readable files, AI text files, markup, or Markdown." No major LLM provider currently consumes llms.txt for ranking. It's a docs-discovery convenience (proposed by Jeremy Howard / Answer.AI in 2024, adopted by Anthropic, Cloudflare, Vercel, Cursor and others for developer docs) — useful, but not a visibility hack.
• **Keyword stuffing.** Measurably negative — −8% in the GEO study. The instinct that helped in 2015 hurts here.
• **Chopping content into tiny fragments.** Google says it's unnecessary; its systems understand multiple topics per page. You're just degrading the reader experience.
• **Ignoring Bing.** It breaks ChatGPT's live search visibility, full stop. Verify in Bing Webmaster Tools.
• **Treating AI search as separate from SEO.** Google's own stance: "Optimizing for generative AI search is… still SEO." Skip the "AEO/GEO hacks," do the fundamentals well.
• **Relying only on owned content.** The citation studies are unanimous — engines lean on third-party and earned media. Your blog alone won't carry it.
• **Weak or inconsistent entity signals.** Inconsistent NAP and a fuzzy brand identity fail the recognition gate before extraction even matters.

How to measure AI visibility

You can't improve what you can't see, and AI answers are non-deterministic — ask the same question twice and the wording shifts. So measurement is a repeatable process on a schedule, not a one-time check. And the metric is not "rank."

Track mentions and share of voice, not position

• **Mention / visibility rate** — the share of your priority prompts where your brand appears at all. If you show up in 40% of 200 prompt runs, that's your number to move.
• **Mentions vs. citations** — a mention is being named; a citation is being used as a linked source. Citation is the stronger authority signal and the one that sends real traffic. Track both.
• **Share of voice** — how often you appear versus a fixed set of competitors across the same prompts. This is the win/lose number for the category. (What share of voice means in AI search.)
• **Per-platform, separately** — you can dominate ChatGPT and be invisible in Perplexity. Measure each engine on its own; a single blended number hides where you're actually losing.

A manual baseline you can run today

Before any tool, you can establish a baseline by hand: take each priority prompt, run it 3–5 times across ChatGPT and Perplexity, and log whether you appeared, your position in the answer, which competitors showed up, and which sources the engine cited. Repeat weekly. The trend — not any single run — is the signal.

There's also a second, higher-confidence signal most tools ignore: your own server logs. When an AI engine reads or cites your site, its crawler hits your pages and its answers send real referral clicks. That first-party traffic is ground truth, not a synthetic sample — though it has to be verified, because crawler user-agents are easy to spoof. SourceWatch measures both sides: the mentions and share of voice across ChatGPT, Perplexity, Gemini and Claude, *and* the real (verified vs. spoofed) AI-crawler and AI-referral traffic landing on your site. There's also an MCP server so you can pull your AI visibility straight into Claude Code.

Where to go next

This guide is the map. Each gate and engine has a deeper playbook of its own:

• How to get your business to show up in AI search — the local/SMB-specific angle on entity signals and NAP.
• How to show up in Google AI Search (AI Overviews) — the Google-specific eligibility rules.
• How to show up in AI search results — the extractability and answer-block deep dive.
• How to get AI to recommend your brand — turning a mention into a recommendation.
• How to rank in ChatGPT — the ChatGPT-and-Bing-specific pillar.
• How to track AI mentions — the measurement pillar, in depth.
• What is AI visibility? — the definitions and the four signals, in one place.