Tag: llm-citations

  • How to rank better in LLMs and AI search

    How to rank better in LLMs and AI search

    For about a decade, SEO meant one thing for most of us: getting found in a list of blue links.

    That work still matters. But here’s what changed. A lot of people now get an answer first, from Google AI Overviews or AI Mode, from ChatGPT, from Gemini, from Perplexity. Only after that do they decide which source to open, if they open one at all.

    So you can rank fine in classic search and still be missing from the answer. Frustrating, right?

    Let’s clear something up first. “Ranking” in a large language model isn’t a position you can buy. It’s closer to this question: when a system needs a source, is your page easy to find, easy to read, and specific enough to quote without twisting what you said?

    Nobody can guarantee you a citation. Anyone who says they can is selling something. What you can do is remove the reasons you’d get skipped.

    That’s what this post is about. I’ll cover entity clarity, writing that’s worth citing, honest structured data, brand mentions, crawler access for AI bots, and a format that works for people and machines.

    What “ranking” in LLMs can reasonably mean

    Google is the clearest official voice on its own AI features. Its documentation on AI features and your website says there are no extra technical requirements to show up in AI Overviews or AI Mode. No special schema either. A page just has to be indexed and eligible to show in Google Search with a snippet.

    The same document repeats the usual SEO basics:

    • Allow crawling.
    • Use internal links.
    • Put important information in text.
    • Keep structured data consistent with the visible page.
    • Keep your Business Profile or Merchant Center info current, if those apply.

    And inclusion still isn’t guaranteed.

    ChatGPT, Gemini (the app), and Perplexity are different animals. Depending on the mode and the question, they may pull live sources, lean on training data, or do both. What you can influence is more down to earth: Can the right crawler reach your page? Is the page specific and clearly about a named entity? Can your claims be quoted accurately? Is your brand referred to the same way around the web?

    If someone promises you a reliable “LLM rank,” don’t take it on faith. I wouldn’t.

    Make the entity obvious

    Systems retrieve and cite more reliably when they can tell who you are, what the page is about, and how it fits with the rest of your site. That’s traditional entity SEO. And it’s become more important, not less.

    Here’s what that looks like day to day. Use one consistent name for your organization, your authors, and the products or services you write about. Your homepage, About page, bylines, and contact info should all agree. If you work in a specific place, say so in plain words, not only in the footer. Put the subject in the title, the H1, and the first couple of sentences.

    Ambiguity costs you. If a system can’t tell whether “Mercury” is a company, a planet, or a metal, it’ll go find a less confusing source.

    Thin, interchangeable posts don’t travel well in AI search either. When I read a page that could’ve been written about any brand in the category, I always ask: what would a model even hold on to here? A page that describes an actual method or a first-hand observation gives it something real. That’s the page that gets pulled.

    Write so a system can cite you without distorting you

    Answer engines like passages they can lift without having to fix the meaning. That changes how you write. What you target is only half of it.

    • Lead with the answer. Open important sections with a direct answer, then add the conditions and detail. A heading that’s a real question, followed by a plain answer, is much easier to retrieve than a slogan.
    • Make claims concrete and bounded. “Meaningful SEO trust often takes six to eight months of persistent work” is something a reader can use. “We deliver explosive growth” isn’t.
    • Credit your sources. If a fact isn’t yours, link to the official or named source. If you can’t name a source, don’t make up a statistic. Ever.
    • Keep the text complete. Google’s AI features guidance says plainly that important content should be available as text. Don’t bury the answer in an image or a tab that never renders as HTML.
    • Date your updates. When the substance changes, update the page and say what changed.

    Bonus: the same habits help your classic snippets and title tags. Two birds.

    Structured data, used honestly

    Structured data doesn’t come with a secret AI-ranking bonus. Google says you don’t need special schema to show up in AI Overviews or AI Mode, and that any structured data you do use should match the visible text on the page.

    So use schema when it accurately describes something already on the page: an Organization, a Person, an Article, a LocalBusiness, a Product, or an FAQ people can actually see.

    Don’t mark up reviews you don’t have. Don’t credit authors who didn’t write the piece. Don’t add FAQ markup for questions that never appear on the page. Inaccurate markup isn’t a shortcut. It’s a trust problem waiting to happen.

    When your Organization name, logo, and official profiles line up in the markup and on the page, you’ve given systems a cleaner identity to attach mentions to. That’s useful. It still doesn’t replace being mentioned and linked as a real source.

    Brand mentions, and the sites that already cite you

    Classic SEO taught us to care about links. They still matter.

    But unlinked mentions and consistent naming help too. They help a model connect a claim to a brand, especially when those mentions show up on sources the system already treats as reliable.

    Can you fake this? Not usefully. A burst of identical bios on low-quality directories won’t make you the go-to source for a technical question.

    What helps, slowly, is the same work that has always built authority. Publish pages other practitioners would actually cite. Be accurate when someone quotes you. Show up in places your buyers already trust.

    If your public footprint is thin, start by making the official record consistent. Then write a few genuinely specific pages instead of a big stack of interchangeable ones.

    Give AI crawlers a clear, deliberate path

    Retrieval requires access. This is the most technical part of the whole job, and it’s the part I see sites get wrong by accident more than any other.

    OpenAI documents its crawlers separately. OAI-SearchBot is used to surface websites in ChatGPT’s search features. Sites that opt out won’t show up in ChatGPT search answers, though they can still appear as navigational links. GPTBot is the one used to crawl content that may go into training generative AI foundation models.

    Those two controls are independent. Keep that in mind.

    ChatGPT-User is a fetcher that kicks in when a user asks for something. It’s not an automatic crawler in the same sense. OpenAI also notes a robots.txt change can take about a day to be reflected, and firewalls can block crawlers even when robots.txt lets them in.

    PerplexityBot is used to surface and link websites in Perplexity results, and according to Perplexity, it isn’t used to crawl content for foundation models. Perplexity-User handles user-initiated fetches and generally ignores robots.txt, since a person asked for the page.

    Google’s Search AI features rely on ordinary Search eligibility. That means Googlebot has to be able to crawl and index the page. Google also documents other product tokens, including Google-Extended, in its crawler overview. Those tokens don’t replace being indexable in Search.

    Here’s the audit I run:

    1. Read your robots.txt. Make sure you haven’t disallowed OAI-SearchBot, PerplexityBot, or Googlebot on the URLs you want visible, unless that’s a deliberate policy.
    2. Check your CDN, WAF, and bot-challenge rules. A 403, a JavaScript challenge, or a login wall can shut out a crawler that robots.txt allows. When a provider publishes IP ranges, use those official lists, not a one-time sample from your logs.
    3. Compare what the crawler sees with what a user sees. The crawler should get the same main content, not a blank shell that only fills in after a client-side app loads.
    4. Decide training access separately from search access. And write the decision down.

    That third one catches more sites than you’d think.

    What not to do

    Don’t buy “submit your site to 50 LLMs” services. Don’t paste a hidden block of machine-only answers at the bottom of every page. There’s no evidence it helps, and it tends to create thin, duplicate content.

    Don’t confuse blocking GPTBot with blocking all visibility in ChatGPT. If you want to appear in ChatGPT search, the crawler that matters is OAI-SearchBot.

    Don’t invent statistics, fake expert quotes, or schema that doesn’t match the page.

    And don’t assume you need a special file or a “GEO score” for Google AI Overviews or AI Mode. Google has said the opposite.

    How this work shows up in a real engagement

    When I work on AI search visibility at Seopya, I start with the same foundation I’d use for classic SEO. Are the important pages indexable? Internally linked? Written with real specificity?

    Then I check crawler access, entity consistency, and whether the pages are actually citable. It all lives inside our SEO and AI search services.

    Here’s the blunt version: if the site can’t be crawled or the content is interchangeable, no amount of “LLM optimization” will make it a preferred source.

    Key takeaways

    • You can’t buy a reliable spot inside ChatGPT, Gemini, Perplexity, or Google AI Overviews. You can make your pages easier to retrieve and safer to quote.
    • For Google AI Overviews and AI Mode, the official requirement is ordinary Search eligibility. No special schema.
    • Be consistent about who you are, write bounded claims, keep structured data in line with visible text, and allow the crawlers that power search visibility.
    • Mentions and citations around the web still help, as long as they’re real.

    FAQ

    Is GEO different from SEO?

    Generative engine optimization is a handy name for the work of becoming a source in AI answers. It doesn’t replace technical SEO, on-page clarity, or earned authority. For its own AI features, Google treats SEO fundamentals as the relevant work.

    Should I block AI crawlers?

    That’s a policy decision, not a default. If you want to be eligible for ChatGPT search or Perplexity results, allow the search crawlers those companies document. Training crawlers can be a separate call. Make the decision on purpose, then check that your security tools actually honor it.

    Does structured data get me cited by ChatGPT or Perplexity?

    There’s no official structured-data requirement that guarantees a citation in those products. Accurate markup can make your entity easier to interpret, especially for Google. It won’t make up for a page a crawler can’t reach, or claims too vague to quote.

    The next step

    Pick five pages that should already be sources. Check each one: Is it indexable? Internally linked? Clear about the entity? Does it give a direct answer? Is the schema consistent (if you use any)? Can the crawlers you want to allow actually reach it?

    Fix those five before you publish ten more.

    If you’d rather have help, get in touch. I can tell you what an AI search pass on your site would involve.

  • E-E-A-T in practice: experience, expertise, and being citable

    E-E-A-T in practice: experience, expertise, and being citable

    Here’s how a lot of people handle E-E-A-T. Add an author box. Mention “years of experience” somewhere. Cross your fingers that Google (and now the AI answer engines) decide the page is trustworthy.

    That’s not how it works.

    Experience, expertise, authoritativeness, and trust have to show up in the work itself. A nice bio can’t rescue a page that makes vague claims, hides who wrote it, or repeats what every other site already said.

    Google talks about this in its guidance on creating helpful, reliable, people-first content, and in more detail in the Search Quality Rater Guidelines, where E-E-A-T is a concept the raters use to judge pages. You can’t look it up in Search Console. There’s no special tag for it. It’s simply a way of asking: should this page be trusted for what it’s trying to do?

    And now there’s a second audience asking that question. ChatGPT, Gemini, Perplexity, and Google AI Overviews all need sources they can quote without twisting the meaning. A page that shows first-hand experience, names its author, and keeps its claims honest is a lot easier to cite than one that sounds confident and says nothing specific.

    So let’s get practical. I’ll show you how to put experience on the page, how to make expertise checkable, and how to write so a system can cite you accurately.

    What E-E-A-T is (and what it isn’t)

    The four letters stand for experience, expertise, authoritativeness, and trust.

    Trust is the big one. The other three are there to support it. A page can be written by a genuine expert and still fall flat if the claims are sloppy, the byline is missing, or the site around it looks careless.

    A few distinctions I’d keep straight:

    • It’s a quality concept, not a plugin metric. Google has said that content showing strong E-E-A-T is useful, especially on topics that can affect someone’s health, finances, or safety. That doesn’t mean any widget can tell you your “E-E-A-T score.” If one claims it can, be skeptical.
    • The extra E (experience), which Google added in late 2022, is about first-hand knowledge. You’ve used the product, done the work, visited the place, or actually been in the situation the page describes.
    • Authoritativeness is often earned off your site. Other reputable sources referring to you, consistently, for the thing you actually do.
    • Trust includes boring hygiene. A real identity, accurate contact details, clear policies where they’re needed, and pages that don’t contradict each other.

    If you remember only one line from this post, make it this one: E-E-A-T is demonstrated, not declared.

    Put experience where a reader can see it

    Experience is the easiest letter to fake in a sentence. It’s the hardest to fake across a whole page.

    “In my experience” followed by a generic tip isn’t experience. A specific observation is.

    On a service or advice page, real experience looks like:

    • The actual sequence you use, including the dull steps and the places where the method stops working.
    • Details someone rewriting a competitor’s article couldn’t have made up: the order you run an audit in, the one field in a form that actually matters, the mistake you see over and over and how you tell it apart from a false alarm.
    • Named, checkable context when you have it. And silence when you don’t. If you can’t publish a metric, don’t invent one. Point to a public case only when the facts are already public, like we do on the Amazon SEO page.

    On a product or “best of” page, experience means evidence that you used the thing: what you tested, under what conditions, and what you wouldn’t recommend. On a local page, it means knowing the place and the service area, beyond a city name dropped into a template.

    My test is simple. Strip out the brand name and the adjectives. Is anything specific left?

    If not, the page hasn’t shown experience yet.

    Make expertise and authorship checkable

    Expertise isn’t a job title in italics. It’s the reader being able to tell who’s responsible for the page, and why that person is in a position to say what they’re saying.

    When I review a site for this, here’s the pass I make:

    • Real bylines. The author should be a person (or a clearly identified team), not a ghost brand account. The exception is when the organization really is the author.
    • An about page that agrees with everything else. Publish an author or about page with the same name, the same role, the same location if that matters, and a short account of the work. Not a résumé stuffed with keywords.
    • Show your homework. When you rely on an official rule, link to the official document. When you’re offering a judgment call, say so, and say what it’s based on.
    • Keep it current. Update the page when the substance changes, and show the date. An expert who never revises a technical article eventually stops looking like one.
    • No phantom reviewers. Don’t attach an “expert reviewer” to a page that person never reviewed. If someone checked the work, say what they checked. If nobody did, don’t make one up.

    The same standard applies to the site around the article. A careful post on a site with no contact info, two different business names, and copy-pasted policy pages will still feel shady.

    Identity is part of expertise.

    Write claims that can be cited without distortion

    This is where old-school E-E-A-T and AI search overlap.

    Answer engines like sentences they can lift without having to fix the meaning. Funny enough, that’s exactly what a careful human reader wants too.

    How to write that way:

    • Say the point plainly, then add the conditions. “A page must be indexed and eligible for a snippet to appear as a supporting link in Google AI Overviews or AI Mode” is citable. It’s bounded, and in that case it’s documented. “Do this to win AI Overviews” is not.
    • Skip the implied guarantees. I won’t promise first-page rankings, citations, or lead volume. I’d be wary of anyone who does.
    • Separate facts, methods, and opinions. A reader (or a model) should be able to tell which is which.
    • Keep important answers in visible text. Not only in an image, a podcast, or a PDF.
    • Match your structured data to the page. If you use schema for an author, article, or organization, it has to reflect what’s visible. Schema that inflates credentials is a trust problem, not a trick.

    In my experience, a citable page is usually just a slightly more careful version of a useful page. It isn’t a different genre of content.

    What not to do

    Don’t buy author bios, fake credentials, or “expert” quotes. Don’t generate a thin article and then dress it up with a stock photo and a made-up reviewer.

    Don’t confuse volume with authority. Ten interchangeable posts on the same topic, each one announcing its own expertise, are weaker than one specific piece a practitioner would actually bookmark.

    Don’t treat an E-E-A-T checklist as the work itself. An author box, a headshot, and a “last updated” stamp are useful when they’re true. They don’t replace first-hand detail, accurate claims, and a site that stands behind its contact information.

    And don’t hide who runs the site. That goes double if you write about money, health, or legal issues. If readers can’t tell who’s responsible, why should they trust the advice?

    How this work shows up in a real engagement

    When I work on E-E-A-T at Seopya, I start with the pages that already have a job: service pages, key articles, the About page. Then I ask whether a skeptical reader could tell who wrote them, what first-hand work they’re built on, and which claims can be checked.

    From there I fix the identity gaps and tighten the writing. Only after that do I worry about how those pages connect to the rest of the site. It all sits inside our SEO and AI search services, because trust isn’t a separate campaign. It’s the quality of the source you’re asking Google and the answer engines to use.

    One more honest note. This kind of trust usually takes persistent work over months, not a weekend of bio updates. I say the same thing about SEO timelines in general, and it holds here too.

    Key takeaways

    • E-E-A-T shows up in the work, not in an author box. It isn’t a score in Search Console.
    • Experience means specific, first-hand detail, not “in my experience” stuck in front of a generic tip.
    • Authorship should be checkable, consistent across the site, and honest about who actually reviewed the page.
    • Bounded claims, linked sources, and answers in plain visible text make you easier to cite.
    • Never invent credentials, metrics, or reviewers.

    FAQ

    Is E-E-A-T a Google ranking factor?

    Google doesn’t present E-E-A-T as a single number you can tune on its own. It’s a concept for thinking about whether content is helpful and reliable, including in the Quality Rater Guidelines.

    Improving the substance, authorship, and accuracy of a page can support visibility. Buying the look of authority won’t.

    Does E-E-A-T help with ChatGPT or Perplexity citations?

    There’s no official “E-E-A-T tag” those systems look for. But a page that’s specific, attributable, and consistent is easier for an answer engine to use as a source than one that’s vague or anonymous. The overlap is real. It’s just not a guarantee.

    Do I need a personal byline on every page?

    No. Not every page needs a personal essay, and some organizational pages are rightly authored by the company. Someone accountable should still be easy to identify, though. And article-style advice is usually stronger with a real person attached, as long as that person actually did the work.

    The next step

    Pick three important pages and read them like a skeptic would.

    Can you tell who’s responsible? Is there any first-hand detail a competitor couldn’t have copied? Are the claims bounded, and when they lean on someone else’s rule, do they link to it?

    Fix those three pages before you roll out a new “trust” template across the whole site.

    If you’d like help with that pass, get in touch. I’m happy to talk through what would really need to change on your pages, and what a byline alone won’t fix.