There’s a question I kept coming back to for months: why does some content show up inside ChatGPT’s answers, while similar content from seemingly stronger sites doesn’t?
Not in the blue links. In the actual generated response — the paragraph ChatGPT writes before you even scroll down to the sources. That’s where the traffic is going. That’s where I wanted to be.
What I found wasn’t exactly what I expected. Some of it is familiar. A lot of it isn’t. And the gap between sites that understand this shift and sites that don’t is widening faster than most people realize.
Semantic relevance — and why keywords stopped mattering the way they used to
Here’s the honest version of this section: I spent years tracking keyword density. There was a spreadsheet. It had a formula. The formula was wrong.
Search engines — and especially the AI layers on top of them — understand what a piece of content is about. Not which words appear 14 times. Google’s BERT update in 2019 was when this really landed for me, because suddenly pages that were “technically optimized” started losing to pages that were just… better written.
The practical shift is covering topics instead of targeting phrases. An article about email deliverability that actually covers SPF records, sender reputation, bounce management, and inbox placement will outperform an article that says “email deliverability tips” in every other sentence. Even if the second article has more backlinks.
Entities matter too. Specific ones. Mentioning Klaviyo, Mailchimp, CAN-SPAM compliance, and double opt-in in context tells an AI system more about your topical authority than any keyword density calculation ever could. These named entities function as signals — they tell the model “this page actually knows what it’s talking about.”
There’s a structural implication here that people miss. If you’re writing about project management, a page that touches on Agile methodology, sprint planning, retrospectives, Kanban boards, and specific tools like Asana or Linear is treated as more authoritative than a page that just says “project management tips” with synonyms rotated through. The entities create a semantic fingerprint.
And honestly — if your content sounds weird when you read it out loud, it’s going to rank like it sounds weird. AI systems were trained on human language. Write human language.
E-E-A-T: the Experience piece is doing more work than most people think
When Google added the second E — Experience — to their quality evaluator guidelines in 2022, a lot of content people treated it as minor. It wasn’t.
The distinction matters because AI can produce competent content. What it genuinely cannot do is tell you what it felt like to lose a major client account, or what actually changed after running the same A/B test for six months, or the specific thing you noticed at a conference that changed how you thought about conversion. That texture of lived experience is something evaluators are specifically trained to look for — and something readers can feel even when they can’t name it.
In concrete terms: use first person when sharing findings. Not “research suggests” — “I tested this across 30 accounts and saw X.” Include author bios with real credentials that are actually verifiable. Cite the original source for any statistic, not some other blog that cited it. Get your name or brand referenced on other authoritative sites. Screenshots of actual results beat generic claims every time.
One thing worth adding here: expertise and experience aren’t the same. You can have expertise (accumulated knowledge) without experience (having actually done the thing). A doctor who has treated a condition has both. A medical writer who researches conditions has only one. In Google’s evaluation framework, the experiential layer is increasingly what separates genuinely useful health, finance, and legal content from competent-but-generic alternatives.
The content landscape is separating into two categories pretty fast — real expertise and AI-assisted filler. The middle is disappearing. Worth knowing which side you’re on.
Technical SEO: still unglamorous, still non-negotiable
Nothing new here, but it keeps needing to be said: great content on a technically broken site doesn’t rank. And now there’s an additional wrinkle — AI search systems pull structured data directly, so technical issues can make you invisible in AI responses even when you’re ranking fine in traditional search. A detailed guide on technical SEO best practices can help you audit and fix the foundations before anything else
Core Web Vitals matter. LCP, CLS, INP — Google uses these as signals, and failing them creates headwinds even if they don’t always disqualify you outright. More practically, slow pages lose readers before they get to content worth citing. A page that takes 4 seconds to load on mobile has already lost a meaningful share of visitors before they see a single word.
Structured data markup (Schema.org) is where I’d spend time if I haven’t already. FAQ schema in particular makes your content easy for AI systems to parse as question-answer pairs, which increases the chances it gets surfaced in AI Overviews—especially for tools like AI add maker that rely on clear, structured answers. Article schema tells crawlers the content type, author, and publication date. HowTo schema works exceptionally well for instructional content. It takes a few hours to implement properly and the upside is real.
Mobile-first indexing has been Google’s default since 2019. Your mobile experience isn’t a secondary consideration — it’s the primary one. If your desktop site is polished but your mobile version is slow or broken, your rankings reflect the mobile version.
The basics — crawlability, internal linking, logical URL structure, HTTPS — these have been true for years. If they’re not handled, handle them before worrying about anything else. A site with broken crawl paths and confusing URL structures loses authority regardless of how good the content is.
One technical element that gets overlooked: page experience signals beyond Core Web Vitals. Intrusive interstitials (especially on mobile), deceptive design patterns, and excessive ad density all generate negative signals. Google’s documentation on page experience is worth reading in full — it covers more than most audits check for.
GEO (Generative Engine Optimization) — the part most content strategies are missing
I came across this concept kind of sideways. I was trying to understand why a particular page I’d written — one I thought was genuinely good — wasn’t showing up in ChatGPT responses about the topic. The traditional SEO metrics all looked fine. Rankings were solid. But in AI-generated answers, nothing.
GEO is the name that’s stuck for this problem: optimizing to be cited by AI systems, not just indexed. It’s a different goal than traditional SEO, and the signals that matter are different too.
The one that surprised me most was how much raw statistics move the needle. Not framed insights — actual numbers. “Email open rates dropped 12% year-over-year” gets cited. “Email engagement has declined” doesn’t. I started going back through older pieces and adding data points almost mechanically, and the difference in how those pages showed up in AI responses was noticeable.
The second thing was sentence structure. AI systems extract verbatim sentences. A sentence that works standalone — that makes a complete, usable point without the three sentences before and after it — is more likely to get pulled into a response. Once I started thinking about this, I realized a lot of my better-performing sections had naturally clean, quotable sentences, while the weaker ones were full of dependent clauses and context that only made sense in sequence.
There’s also something about where your answer appears in the document. I’ve noticed — and research backs this up — that AI systems tend to pull heavily from the beginning of a page. Not exclusively, but disproportionately. If the actual answer to someone’s question is in paragraph 14, you’re competing at a disadvantage against a page that puts the answer in paragraph 2 and elaborates from there.
One more thing worth knowing: Perplexity, ChatGPT, Google AI Overviews, and Bing Copilot don’t behave identically. Perplexity tends to cite more sources and pulls from newer content more aggressively. Google AI Overviews lean toward pages that already rank. If you know which platform your audience actually uses, you can weight your optimization accordingly.
For getting a read on where a specific page stands, WriteSEO has a free checker at writeseo.vercel.app/geo-check — paste any URL and it breaks down the signals. I’ve used it enough to trust the output.
Content freshness — and what “updating” actually means
AI systems with live web access weight recency. A guide from 2019 that hasn’t been touched competes poorly against a 2025 update on the same topic, even with a stronger backlink profile.
The fix isn’t complicated but it requires consistency: display “Last Updated” visibly on the page (not just in metadata), add actual new content when you update (not just a date change — crawlers notice), and review evergreen content quarterly. In fast-moving fields especially, six-month-old statistics can make an otherwise solid piece feel dated.
What most people miss is the difference between updating and re-dating. Changing the publication date without adding substance is something Google has gotten better at detecting. The content hash changes, yes, but if the substantive content didn’t change, the freshness signal is minimal. What actually works is adding new data, updating statistics that have changed, adding a new section on a development since the original publication, or revising recommendations based on current tools and practices.
There’s also a strategic angle here. Some topics have natural update cycles — annual reports, tool comparisons, regulatory changes. Building a content calendar that revisits these pages at the right intervals, rather than letting them drift, keeps them competitive without requiring entirely new content creation.
For AI-focused content especially, the pace of change is fast enough that something written 12 months ago about specific AI capabilities or products may be substantively wrong today. Accuracy isn’t just a quality issue — it affects whether AI systems will cite you, because citations to outdated or incorrect information create trust problems for the AI systems themselves.
Links: still a core signal, still worth doing right
The fundamentals of link authority haven’t changed. Editorial links from topically relevant, authoritative sources pass real value. Buying links from directories passes almost nothing. That’s been true for a while, and the algorithmic ability to distinguish genuine editorial links from manufactured ones keeps improving.
What’s worth emphasizing is that link relevance has become more important relative to raw link count. A link from a well-regarded marketing publication to your marketing content is worth more than ten links from general-purpose sites in unrelated niches. The context in which a link appears — the surrounding text, the page topic, the linking domain’s authority in your specific niche — all factor in.
Digital PR — actually getting cited by journalists, mentioned in industry research, quoted in publications — creates a different category of link than anything manufactured. These links come with surrounding context that reinforces topical relevance, from domains with established authority, in content that’s likely to attract further links. Not scalable in the traditional sense, but uniquely resistant to algorithm updates because they’re structurally different from anything you can buy.
Guest posting works when done right. The key qualifiers: the host site should be topically relevant, have a genuine audience, and maintain editorial standards. Guest posts on sites that clearly exist to host guest posts (thin content, no real audience, inconsistent topic coverage) pass diminishing value.
Internal linking is consistently underused. If you publish a strong comprehensive guide, link to it from every relevant post you have. Build topic clusters with clear hub-and-spoke structure — a main pillar page that covers a topic broadly, with supporting cluster pages that go deep on specific subtopics, all linking back to the pillar and cross-linking to each other. This structure helps both crawlers understand your topical coverage and AI systems understand how your content relates.
Anchor text still matters, though it’s easy to overthink. The simple version: describe what you’re linking to in plain language. “How I fixed our crawl coverage issue” is better than “click here.” Exact-match anchors repeated across many links to the same page can look off — I’ve seen this create problems even when it wasn’t intentional — so vary the language while keeping it descriptive.
Behavioral signals: the part you can’t fake at scale
I used to think of behavioral signals as a consequence of good SEO — something that happened naturally once your rankings improved. I’ve changed my mind on this. Behavioral signals feed back into rankings, so they’re worth thinking about proactively.
The most actionable one is click-through rate. Your title tag and meta description are what people see before they decide whether to visit your page. I’ve had pages that ranked #4 and consistently outclicked #1 because the title was sharper and more specific. Over time, those pages moved up. The description matters too — it’s not a keyword stuffing field, it’s a 160-character pitch for why someone should visit. Most sites treat it like a formality.
Dwell time is trickier because Google doesn’t surface it directly, but you can feel it in the data. When a page has high bounce rate combined with very short sessions, that’s users coming in and immediately deciding it wasn’t what they needed. The fix isn’t technical — it’s making sure the page delivers on what the title promised. Obvious in theory, harder in practice when you’re writing about things that are genuinely complex.
Return visits are something I’ve come to care about more than I used to. They signal that someone found enough value to come back, which is a qualitatively different signal from a one-time visit. Brand searches — when someone types your site name directly — are even stronger. These aren’t things you can manufacture. They’re the result of publishing things people actually want to read again.
Social sharing matters indirectly. The direct effect on Google rankings is debated, but the downstream effects — more backlinks, more direct traffic, brand mentions that AI systems pick up — are real.
Putting it all together
None of these elements work in isolation. I’ve seen technically flawless sites with weak content and strong-content sites with broken crawl paths — neither performs as well as sites that get both right.
What’s changed in the last few years is the addition of AI citation as a goal alongside traditional rankings. These aren’t competing objectives. A page optimized for GEO — with clear structure, specific data, quotable sentences, and authoritative sources — is also a better page by most traditional SEO measures. The skills transfer.
The sites that seem to be pulling ahead right now are treating AI search as a distribution channel to understand, not just a tool to use for writing. There’s a difference between asking ChatGPT to write your content and asking yourself what it would take for ChatGPT to cite your content. The second question is more interesting and more productive.
Where I’d start, practically: a technical audit first, because unfixed crawl issues are the kind of thing that quietly undermine everything else. Then a real look at topical coverage — not keyword coverage, but whether you’ve actually written the most useful thing on the internet about your specific niche. Then E-E-A-T: is there a real person behind this content, with real credentials, making real first-person claims? And finally, GEO: are the pieces structured so an AI system could actually pull and cite them?
It’s the same craft as always. Just with a few more things to pay attention to.
Jeez builds WriteSEO, a suite of free SEO and GEO optimization tools for content creators and small businesses. Free GEO checker: writeseo.vercel.app/geo-check