How to appear in ChatGPT, Perplexity, and Google AI Overviews in 2026 (no fluff)

Search «how to appear in ChatGPT» on Google and every result says the same thing: index on Bing, add Schema, write useful content. True and insufficient. I've been auditing what sites actually get cited in ChatGPT, Perplexity, and Google AI Overviews responses for real clients these past months, and the patterns that work aren't the ones everyone repeats.

This isn't a generic checklist. It's what I saw when measuring, what actually weighs, and what looks important but moves nothing.

Three engines, three logics

Treating «SEO for AI» as one thing is the first mistake. These three consume information differently:

• ChatGPT (OpenAI): internal search based on Bing + own web crawl (GPTBot). Citations typically appear in «search» mode. Sources per response: usually 3-8.
• Perplexity: own crawler (PerplexityBot) + near-instant indexing of «trending» pages. Almost always cites 5-12 sources, preferring recent content with topical authority.
• Google AI Overviews: served from Google's index as-is. Pages cited are the same ones ranking top 10 organic for the query, filtered by «extractability» — what paragraph extracts cleanly.

Practical consequence: optimising for all three at once is possible, but each has one lever that weighs more.

Engine	Critical lever	Secondary lever
ChatGPT	Bing indexing + structured data	Measurable topical authority
Perplexity	Freshness + answer clarity	Article and FAQPage Schema
Google AI Overviews	Already-consolidated top 10 organic	«Extractable» paragraphs (subject-verb-direct answer)

What DOES work (measured)

1. Bing Webmaster Tools + IndexNow

Basic, not optional, and almost nobody does it right. For clients depending on ChatGPT:

1. Register the property in Bing Webmaster Tools.
2. Submit the sitemap (same XML you submit to Google Search Console).
3. Activate IndexNow. WordPress plugins exist (manual or automatic), or implement with 30 lines of code in Next.js / Astro. IndexNow is an API that pings Bing (and Yandex) as soon as you publish or update a URL. Without it, Bing can take 2-3 weeks to crawl a change. With it, indexing typically in 24-48 hours.

In 8 of 10 SMB sites I audit, this isn't there. Without it, ChatGPT won't cite you no matter how good your content is.

2. Specific Schema markup, not generic

«Add Schema» is like saying «eat healthy». Too vague. What moves the needle:

• FAQPage on any page with real questions. ChatGPT and AI Overviews love Q&A format because it's directly extractable. If your services page has 6 real client questions marked as FAQPage, citation chances rise.
• Article or BlogPosting with real author, datePublished, dateModified. Freshness weighs more than it seems.
• HowTo for step-by-step tutorials. Still works even though Google removed its rich snippet — AIs still read the markup.
• Person or Organization with sameAs pointing to real LinkedIn, GitHub, Twitter/X. AIs build «topical authority» by triangulating entities.

Not generic WebPage. Not Thing. Specific.

3. Machine-readable content structure

AIs don't read your fancy CSS or animated hero. They read post-render HTML. What helps:

• Question as H2 or H3 + direct answer in the first paragraph below. Example: H2 «How much does an SEO audit cost?» followed immediately by «A full SEO audit costs between €1,500 and €5,000 depending on site size.» That's an extractable paragraph.
• Short list items. No 4-line bullets with subordinate clauses. 8-15 word bullets each.
• Tables for comparisons. This page has one above — not for aesthetics, but because tables are the easiest structure to cite verbatim.
• TL;DR at the start of the post. AIs almost always identify it as the main answer.

4. Measurable freshness

AIs penalise old content more than assumed. Indicators they DO read:

• <meta property="article:published_time">.
• Schema datePublished and dateModified.
• Year in title («Local SEO in 2026» appears more in Perplexity than «Local SEO: definitive guide»).
• Whether the post is actually updated — not fake refresh: changing 3 words and bumping dateModified counts as «date manipulation» and some parsers detect it.

5. Topical authority (not domain authority)

This is the most counterintuitive. The «domain authority» Ahrefs / Semrush reports matters little for AI citations. What matters is topical authority: how many niche sites link to you, how many mentions you have on sector-reference sites.

For a law firm: the local Bar Association mentioning you is worth more than 50 generic blog backlinks. For a dentist: dental societies citing you weighs more than high rankings in generic health directories.

What DOESN'T work (measured)

1. Spamming everything with fake FAQs

I've seen sites adding 30 FAQs per page, all with the same repetitive structure, marked as FAQPage. Measured result: zero AI citation impact, and in some cases models downgrade the page as «overoptimisation signal».

Rule: 5-10 real FAQs per service page. The ones clients actually ask. If you don't have 5 real, 3 real beats 30 invented.

2. «Optimising for AI» by rewriting EVERYTHING as bullets

Turning good text into flat bullet lists «so AI reads better» usually backfires: you lose context, AIs detect forced structure, and lower perceived reliability. AIs prefer scannable content, yes, but also need prose to understand context and nuance.

Right balance: H2 + prose paragraph (3-5 lines) + bullets or table when they add real structure, not by default.

3. Buying «AI mentions» from agencies

There are agencies selling «guaranteed citations in ChatGPT». Doesn't work. ChatGPT has no payable mentions system, no backdoor. What those agencies usually do is publish your brand on spam sites for the crawler to index, which sometimes appears in a specific response and often generates negative associations the models detect.

4. Optimising the title without optimising the first paragraph

The title still matters (for Bing and Google to index well). But AIs almost never «cite verbatim» from it. What they cite verbatim is the first paragraph after H1. If your first paragraph doesn't answer the question the title raises, the title is wasted.

5. Generic Schema like «WebPage»

Marking everything as WebPage is like marking nothing. AIs extract no value from it. Specific Schema (Article, FAQPage, Service, HowTo, Recipe, Event…) or no Schema.

How to measure if you're being cited

This is the hardest and almost nobody does it. Three ways that work:

1. Brave Search API + source filters. Brave lets you query which domains appear cited in its AI response (Brave Summarizer). Free API up to a certain volume.
2. DataForSEO «LLM Mentions» API. Paid, but gives aggregated mentions in ChatGPT and Google AI Overviews per keyword. Useful for quarterly benchmarks.
3. Disciplined manual search. Every month, your 10 most important queries, in ChatGPT (search mode), Perplexity, and Google. Note in a spreadsheet: do they cite you? in what position of the source list? with what snippet?

Way 3 gives the most insight because you see the exact context where you appear.

The 90-day plan

If you're starting from zero (site not appearing in any AI yet):

Day	Action
1-3	Register Bing Webmaster Tools + submit sitemap + activate IndexNow
4-7	Audit current Schema with Schema Validator (not Rich Results Test). Identify pages without specific type
8-21	Implement specific Schema (FAQPage, Article, Service) on top 10 most important pages
22-30	Restructure first paragraph of each core page with direct answer to the title's question
31-60	Create real FAQ block (5-10 questions) on service pages. Get 2-3 niche site mentions
61-90	Measure baseline AI citations (manual + Brave API). Iterate on pages that do appear, to appear in more queries

At 90 days, a well-worked SMB site starts appearing in at least 1 of the 3 engines for its core queries. At 180 days, usually appears in all 3 if the sector isn't saturated with big brands.

Where to start tomorrow

If you have 30 minutes today and want to move one single thing that does weigh:

1. Open Bing Webmaster Tools.
2. Verify the domain.
3. Submit the sitemap.
4. Activate IndexNow (WordPress has a plugin, other stacks 30 lines of code).

That alone, with no other changes, in 4-6 weeks improves how often ChatGPT cites you. Not magic, just the first link in the chain almost nobody closes.

---

A full AI visibility audit is included in my SEO audit service or as an added module in monthly technical SEO. Contact for a concrete proposal.