Google picks, AI cites. SEO works so it’s you

Ask a developer what SEO is and they’ll talk about clean code, load times, and pages Googlebot can crawl. A copywriter will point you toward keywords and the quality of your writing, while a PR manager will bring up media mentions, links, and brand reputation. Then someone selling GEO services shows up, and to them the rest is ancient history: these days, they’ll argue, all that matters is getting cited by ChatGPT and showing up in AI Overviews. Each one describes their own piece of the puzzle accurately.

Meanwhile, your domain gets evaluated on several fronts at once. Google builds summaries at the top of the results page, and for some questions searchers get what they need without opening anything; some queries go to a conversational assistant instead of a search engine; and in what these systems choose, the brands the web recognizes and cites carry weight. When SEO gets split into pieces, each person improves their own piece, the numbers stay flat, and the budget gets spent without anyone knowing exactly where.

Technical work, content, reputation, and presence in generated answers all come down to the same operational problem: figuring out why a source gets chosen, where it shows up, and what’s stopping your site from being that source. The outcome shows up in different places – rankings, citations inside answers, searches that include your name, calls from a listing nobody ever opened – and each of those places gets measured with its own tool, speaking a different language from the rest.

What SEO is

SEO, short for Search Engine Optimization, is the work that creates the conditions for a site, a piece of content, a brand, or another digital presence to be found, understood, evaluated, and chosen by search systems when they answer a need someone has. In Italian, «il SEO» refers to the professional who does this work, the SEO specialist.

When you search on Google, the result can be a page, a map, a video, or a direct answer; when you ask the same question to an AI assistant, some of those same sources end up folded into text generated on the spot. In both cases, a system has already decided which resources to consider and which to leave out.

AI answers add another layer on top of that selection process. Inside Google, AI Overview and AI Mode rely on Search’s underlying systems. Outside Google, when an assistant needs to pull in current information, its external search often starts with traditional search engines, Google chief among them, and returns pages those engines have already indexed and ranked. In those cases, the assistant picks from material that organic search has already filtered: if your page is missing from the index and from the results of the searches the system runs, it’s out of the running as a source before the answer even gets built.

Technical, on-page, off-page, and local SEO split up a set of skills that come back together in the outcome. Each one works on a different stage of the selection process, and you notice which stage is missing coverage when something breaks down.

What has to happen	Where SEO operates	What you see when it fails
The engine finds the resource and can process it	Technical SEO: architecture, internal links, crawling, indexing, canonical version, JavaScript-rendered pages	The page exists, but Google doesn’t keep it in the index
The engine matches it to the right query	On-page: keyword research, intent, content, information architecture	You show up for searches far from the ones you actually care about
The engine considers it a trustworthy source	Off-page: links, mentions, digital PR, reputation	You answer the question but lose out to stronger sources
The engine surfaces it for location-based searches	Local SEO: Business Profile, local information, reviews, and signals tied to geographic presence	People searching in your area pick a different business on the map

The same page can pass through all these layers. A technical blocker stops it before the content ever gets evaluated; an editorial answer built on the wrong need clears the technical stage and loses out later; a relevant source competes with a reputation and track record different from its competitors’.

What stayed the same while the results changed

In the mid-1990s, when the history of SEO began, the work was new enough to go by several names: search engine optimization, search engine placement, positioning. The visible result was mostly a list of pages. Today the same search results page mixes videos, maps, products, forums, generated answers, and organic links, yet we still use the same acronym.

Since launch, Google has ranked results partly through PageRank, which reads the link structure between documents as a signal of importance: a page linked from other important pages gains weight of its own. PageRank still contributes to ranking alongside many other signals, and over time language understanding, freshness, originality, quality, and systems built for specific needs have joined it. Lists of 200 ranking factors try to bring order to this complexity, as long as you don’t read them as a recipe that weighs the same for every query.

On the other side of the screen, the way people ask questions has changed. A few words typed into a search box turned into full questions, then long queries with constraints, context, and comparisons. The query got longer; the need behind it stayed the same.

Figuring out what the person is actually looking for and deciding who the engine can trust to answer them: these two problems have outlasted multiple shifts in technique, interface, and format, with SEO working on them the whole time. E-E-A-T, reputation signals, and the weight given to brands are just the latest developments in a question about trust that PageRank already raised through links.

Even in Google’s generative features, AI Overview and AI Mode, Search’s core ranking and quality systems stay active, and the work needed to show up there is still SEO.

Organic, SEA, and SEM

Organic results and ads can share the same screen while getting there through very different logic. The organic result depends on the engine’s selection process. Advertising runs on an auction: budget, bid, targeting, and campaign quality determine access to ad space.

Paying doesn’t make Google crawl your site more often, and it doesn’t buy you a better ranking in organic results. SEA, Search Engine Advertising, works on the spaces you can actually purchase.

The distinction changes what your budget actually buys. With SEA you pay for access to a space under the campaign’s terms, and once the campaign stops, that space goes back up for auction. With SEO you work on the conditions that lead the engine to pick your resource: pages, structure, content, and reputation stay with the project even after the work is done, and keep competing in the selection process as long as they hold up against the competition.

The term SEM, Search Engine Marketing, causes more confusion because its meaning has shifted with usage. It started as a broader umbrella for both SEO and search advertising, but today it’s often used to mean paid search specifically. When you see it in a proposal, the list of deliverables matters more than the acronym itself.

Before ranking, Google has to be able to use the page

Your page loads fine in the browser, the content is live, and the links work. Google Search Console can still tell a different story: URL unknown, resource crawled but not indexed, a different URL picked by Google as the canonical version. To you, the page is published. To Google, it may not have cleared the first stage yet.

Search moves through distinct stages. First it discovers and retrieves resources, then decides which ones to keep in the index; only after that can it compare them once a query comes in. A page stuck in crawling has a different problem from an indexed page that just isn’t showing up for the searches you expected.

Technical SEO works mainly in this zone. Architecture, internal links, server responses, crawl directives and indexing, how the engine processes JavaScript-rendered pages, duplication, and canonical version selection all affect whether the engine can reach and correctly interpret your resources.

When the problem repeats across dozens or hundreds of URLs on your site, checking one page at a time hides the pattern. SEO Spider follows your site’s internal links the way a crawler would, the program a search engine uses to discover and download pages, and shows you which URLs turn out to be reachable during the crawl, along with server response codes, directives, and other technical elements collected along the way. The spider’s crawl describes the structure it can traverse; what Googlebot actually does belongs to the search engine’s own data.

Google’s AI features start from the same technical foundation. A page can appear as a supporting link in AI Overviews and AI Mode when it’s indexed and eligible to appear with a snippet, the text excerpt that accompanies the result. The same technical requirements that apply to Search apply to generated answers. In Search Console, inclusion in generative AI features is on by default, and you can exclude your site using the dedicated control.

After indexing, a different problem starts

When Search Console shows “Crawled – currently not indexed,” you know where Google stopped. The why is something you need to look for in the content, duplication, structure, and directives, along with the URL’s concrete characteristics.

Once a page enters the index, that problem is behind you, and another begins. Google now has a resource it can consider, but the query puts it in competition with everything else the engine considers suitable for that question.

Rankings emerge from that comparison. Query understanding, relevance, links, freshness, quality, and other signals come into play with different weight depending on the type of search: freshness matters one way for news, and another way for a search looking for a stable definition.

The query decides who you’re really competing against

Say you have a commercial page that’s well built, technically accessible, and perfectly aligned with your service. The keyword you want to rank for returns guides, comparison sites, and videos.

You might be bringing a good answer to a different contest than the one Google is actually running.

A query is the measurable part of the question. Behind that string sit the need, the moment of decision, the constraints, the alternatives, and the format the person expects to use to deal with the problem.

The SERP, the results page Google builds for that search, reveals at least part of that interpretation. If Google puts guides, comparison sites, and videos at the top, your real competitor on that query is no longer necessarily the company selling the same product as you. It could be a publisher, a marketplace, a community, or an informational resource that catches the demand long before any offer does.

Keyword research exists precisely to map out that space. Different phrasings can converge on the same answer; others, nearly identical in wording, produce results different enough to need separate resources. Compare SERP gives you a useful signal by measuring how much the results for different keywords overlap.

A high-volume query can open onto a SERP far removed from what you sell. A narrower search might have less demand but a much more direct connection to what you offer.

When one question leads to others

A long request in AI Mode is a single question for you. For the system, it can turn into a series of smaller searches, launched to gather the pieces needed to build the answer.

Google calls this mechanism query fan-out. A source can end up in the answer through a search the user never actually typed, because that query is generated during information retrieval.

The distance from the exact keyword grows, but the connection to the question remains. One of your pages covering a sub-problem in enough depth can still prove useful even when the original prompt contains many other constraints and topics.

Trying to turn every possible subquery into a new page leads toward something Google has already addressed in its own policies: mass-producing content to capture variants and fan-out queries can fall under scaled content abuse, the large-scale production of pages built mainly to manipulate rankings, with little or no value for readers.

What sets apart pages that are relevant to the same search

You search your query and find a SERP full of correct pages, including yours. They all answer the question, all have a readable structure, all cover the expected topics. At that point, relevance stops being what sets them apart.

The difference comes from what a source adds on top: direct experience, proprietary data, evidence, analysis, a primary source, a reasoned position. In generated answers, the system compares multiple sources, and a unique point of view is what makes one stand out. Even when AI enters the production process, the standard stays the same: what matters is what the page gives the reader, regardless of the tool used to write it.

Your page can cover the intent perfectly and still be an interchangeable answer if it just repackages information already available everywhere. Experience, data, and observations that only you have are what change the material the system finds when it compares sources.

A page shows up in the SERP alongside the source that published it

You have a page that’s accurate, up to date, and relevant, and you watch it lose the comparison to another page that says nothing more. Every page enters that comparison along with the subject and context Google can piece together around the source: behind one there might be a recognizable entity, linked to consistent sources and already associated with that topic; for the other, the engine might be able to reconstruct much less.

E-E-A-T helps explain this dimension. Experience, expertise, authoritativeness, and trustworthiness describe characteristics that systems try to recognize through combinations of signals; E-E-A-T isn’t a single ranking factor, and it carries even more weight on topics that can seriously affect people’s lives.

A source builds trust through relationships, mentions, experience, reputation, and identity that the content of a single URL can explain only in part.

When an industry publication mentions your brand and links to your site, the relationship between the two resources becomes explicit: that’s a backlink. A mention without a link still creates an association between the brand and the context; reviews add the experience of someone who has actually dealt with your business. Digital PR works on these presences when it manages to place the brand within sources relevant to your market and to the topics you want to be recognized for.

Where brand fits into SEO work

On your own site, you describe exactly who you are, what you sell, and how you want to be perceived. Ask the same question to an AI system, though, and you may find that part of the description comes from other sources, or that your site doesn’t appear among those used at all.

In our study on how AI systems describe brands, we found exactly this gap: the identity a brand states about itself and the one a system reconstructs can rest on different sources.

The problem starts well before the generated answer. Google also organizes people, companies, products, places, and topics as entities and relationships; the Knowledge Graph makes this logic visible. Name, products, people, locations, and topical associations form a context that the engine tries to reconstruct from multiple sources.

A brand becomes an area of SEO work because how the engine recognizes a source and connects it to a topic depends on that reconstruction. On your own site, you control part of the representation; other sources can confirm it, expand it, or shift it.

In generative models, part of the information about a brand belongs to knowledge the model has already acquired; another part can come from live search. In the model’s memory we see the representation that has become established, not the individual sources that shaped it; in live search, by contrast, the retrieved sources become visible, along with the relationship between those sources and the description produced.

When the chosen resource isn’t on your domain

For many searches, Google surfaces a YouTube video, a forum discussion, a post, or a business listing right up top. These are organic results: their presence there isn’t an ad, and the engine selected them as the best answer to that query.

To be found on Google, you don’t need a website. A professional with no domain can rank on page one with a video the system considers relevant to that search; a discussion started in a community can become the source the engine surfaces to someone looking for another person’s experience before making a choice.

Your domain remains the one place you fully control: structure, content, paths, and the point of conversion all depend on you, and it’s the official reference point that many other sources link back to. But it’s one of several places where your brand gets found, mentioned, and chosen. SEO work follows the resource the moment it enters a search system: a video when it competes in YouTube searches or Google SERPs, a discussion when it gets retrieved for a query. Until that point, it belongs to editorial or community work.

Google’s generative features can surface what’s being said about products and services on blogs, videos, and discussion threads. When a system reconstructs your offering, reviews, threads, and videos can end up in the answer alongside your official pages, and brand visibility takes shape wherever someone looks for it.

A local search can end before anyone reaches your site

The Business Profile brings together hours, reviews, photos, services, directions, and contact details. For many local searches, the path to a decision can end right there.

Google builds local results based on relevance, distance, and prominence. Local SEO work involves the website together with the Business Profile and the other information Google uses to reconstruct local presence and reputation.

When someone calls directly from your listing, the contact originates from the search, and no session shows up in your Google Analytics.

Social, video, and UGC can compete for the same query

YouTube, TikTok, and other platforms have their own search; some content hosted there can also show up in Google results.

Forums, videos, posts, and user-generated content gain the most ground where demand is looking for firsthand experience, opinions, comparisons, or demonstrations. Creators and communities can take up organic space on the very searches where you’re trying to gain visibility.

Social Opportunity starts from the competitor keywords set up in your SEOZoom Project and identifies the ones where social platforms and UGC sites rank among the top results. This data shows you a slice of demand that’s already flowing through resources outside the domains you usually consider competitors.

Search beyond Google adds another layer: those same platforms have their own queries and their own selection criteria. A piece of content can compete within Google, within the platform, or in both environments.

From ranking to getting cited in AI answers

When you read a generated answer, the cited sources don’t necessarily match the top organic results for the same search on Google. Ranking and citation follow different logics.

Inside Google, ranking and citation remain tied to the same systems that power Search. Outside Google, passing through organic search doesn’t end the selection process: when external search starts from traditional search engines, the assistant can re-read the retrieved sources and reorder them based on the question it’s trying to answer. A page that ranks lower in the SERP can turn out to be the most useful one for a specific part of the response.

In the SEOZoom framework, SEO for AI covers the work that goes beyond the scope of Google Search: model memory (what a system learned about a brand during training) and assistants that don’t depend on Google Search. For the portion that relies on live search across traditional engines, organic visibility is the entry condition: a source that those engines don’t show for the searches the system runs never reaches retrieval. For the portion that depends on the model’s memory, the relationships and confirmations spread across the web form the context through which a brand becomes recognizable beyond what it states on its own site, and the model may have encountered them during training.

Getting cited is a new outcome you need to manage. Selling it as a job that can stand on its own starts from the wrong end: technical setup, content, demand, sources, and reputation are the conditions that make a resource available and credible before a generated response ever considers it.

SEO shortcuts and AI shortcuts

Link schemes have spent years trying to artificially build relationships that should have emerged between real resources. Large-scale spam multiplies pages to take up more space. Reputation abuse exploits a host domain’s signals through third-party content published with little or no direct oversight, with the goal of manipulating rankings.

Anti-spam policies and black hat SEO have told this story for a long time: getting the outcome of selection without actually building what that outcome is supposed to represent.

Generative responses have changed the tactics sold as shortcuts, but not the core problem. Artificial text fragmentation, files presented as requirements for models, structured data “for AI,” and mention packages all promise direct access to citation.

For Google, many of these supposed requirements already have a documented limitation: generative search doesn’t require artificial chunking, files like llms.txt, specific structured data, or purpose-built mentions. On other systems the architecture can differ, and so can what you’re able to verify.

Where GEO and AEO diverge

Your brand can be known correctly by the model and still disappear from live responses. The opposite can happen too: retrieval surfaces recent, relevant pages, while the model’s established knowledge keeps linking the brand to old or incomplete information.

In the SEOZoom framework, this gap separates GEO from AEO. GEO concerns what models know and associate with the brand; AEO looks at what happens when the system builds a response at the moment of the query.

GEO Audit lets you read the identity, associations, and semantic relevance attributed to the domain. AEO Audit instead looks at recognition, presence, and consistency in the responses produced by the main AI engines. The tools address different problems, even when the initial symptom looks the same: “the brand doesn’t show up.”

When you need to know whether the issue persists, time adds another dimension. AI Prompt Tracker follows the prompts you choose to monitor and makes citations, mentions, and recommendations comparable across successive checks. AI Visibility then broadens the view to engines, prompts, keywords, cited URLs, and competitors, keeping results in one picture that a single organic position could never capture.

What SEO is for and how you measure its value

SEO puts your business in front of the questions people ask when they have a need: understanding a problem, comparing solutions, choosing a vendor, finding a nearby store. The value lies in access to those moments: pages and content built around a query can cover every search where the engine considers them a good fit, while brand and reputation support the choice at the point where the person compares alternatives.

When you put SEO into a budget, the easiest return to recognize is the visit that turns into a quote request or a purchase. Other parts of the value appear before the site even comes into play, on surfaces where the choice takes shape without a click, or build up over time, when a source encountered more than once already feels familiar by the time the decision gets made.

Between a ranking position and revenue sit the decision to click, the on-site experience, the offer, the price, and the competition: rankings, traffic, and revenue measure different points along the same chain. If you want to attribute revenue to a single keyword or a single action, you need connected data and an awareness of what that data can’t show you.

When ranking position alone no longer explains the result

You look at the clicks chart and watch it drop while the average position barely moves. In between, though, the SERP changed: a response may have gotten longer, a video block may have appeared, local results may have expanded, or part of the query may now get answered before the result even opens.

You might be tempted to tweak the title tag, one of the sources Google draws from to build the title shown in the result, or to blame the drop on an AI Overview effect. The drop alone doesn’t tell you which explanation is right.

SEO rankings still describe the comparison between organic results. They describe much less of what space and attention remain available around that position. A SERP with organic links up front creates a different path than a page where the answer is already anticipated by maps, videos, products, or summaries.

Zero-click searches are the clearest sign of this: part of the need gets met right on the search surface. Visiting the site is still significant, but it no longer equals the full visibility that SEO produces.

Real data and estimates

Impressions in Search Console grow while your Analytics numbers stay flat, or the opposite happens. The two platforms watch different parts of the same path, and a gap between the numbers can simply come from what each one is able to see.

Search Console knows the relationship between the site and Google Search: query, impressions, clicks, CTR (the share of impressions that turn into clicks), average position, indexing status. AI Overview and AI Mode remain included under the Web search type; the dedicated report for generative features also lets you isolate organic impressions earned in those experiences.

Inside your SEOZoom Project, Period Comparison puts the same Search Console data side by side across two timeframes and shows you which piece of performance moved. A drop in clicks alongside stable impressions tells a different story than a drop where exposure disappears too.

Analytics starts after the visitor lands on the site. Sessions, paths, conversions, and revenue belong to that stretch, while query and Search visibility stay inside Search Console. Query and on-site behavior remain in separate sources, so a full reconstruction of the path to a single sale stays incomplete.

The same holds when the comparison goes beyond your own domain. On your site you have direct first-party data; for competitors you can only work with estimates and models. SEOZoom holds both families of data together without changing what they are.

Zoom Authority, for example, is a proprietary metric that summarizes a domain within the SEOZoom model, making a comparison possible even without access to a competitor’s internal data. Search Console, by contrast, tells you what Google actually recorded for the connected property.

The question	First-party data	External observations and estimates
Which searches do I show up for, and how often am I chosen?	Search Console (Google): queries, impressions, clicks, CTR, average position	The makeup of the SERP, showing what’s occupying the space above your result
Why did clicks drop?	Period Comparison (SEOZoom, based on Search Console data): which performance component moved	SERP changes on the same queries
What happens after the click?	Google Analytics: sessions, paths, conversions, revenue	Only your own data shows this
How visible am I in generative experiences?	Generative AI report in Search Console (Google): impressions of your site’s links in AI Overview and AI Mode	AI Prompt Tracker and AI Visibility (SEOZoom): citations, mentions, and recommendations on tracked prompts
How do I compare my domain’s estimated strength against competitors?	Just your own side of the comparison	Zoom Authority (SEOZoom): a proprietary metric for comparing domains without access to their internal data

How SEO gets done

In practice, the work rarely starts from a technique. It starts from something that doesn’t add up: an important page of yours stays out of the index, queries bring you a different audience than expected, the site loses clicks without losing visibility, a competitor shows up in AI answers where your brand is still missing.

The same symptom can mean something completely different once you add a second data point. A page out of the index raises an issue that comes before ranking; a page indexed for the wrong queries shifts the focus to intent and architecture; stable organic presence with falling clicks points you to the SERP; a brand that ranks well on Google but is absent from monitored prompts opens up yet another front.

The hypothesis comes from cross-referencing the data: the page is out of the index because of a directive, it answers a different intent than the one the SERP rewards, it loses out to stronger sources. The fix only tests the hypothesis if it leaves a readable signal: a status change in Search Console, movement on the queries involved, a citation that appears or disappears in monitored prompts. Inside your SEOZoom Project, the AI SEO Assistant ranks suggestions by priority starting from your project’s data, so different problems produce different recommendations.

Then the signal shows up, or it doesn’t, and the diagnosis gets updated. A fix with no effect narrows down the hypotheses once enough time has passed, the change has been picked up, and the chosen signal was the right one; before that point, it only means the result isn’t readable yet. An effect that shows up on queries other than the ones you expected opens up a new reading. SEO work repeats this cycle on whatever is limiting the project at that moment, while competitors, SERPs, and demand keep moving in the same period.

Who does SEO, and what they can promise

Generative AI optimization now falls among the services an SEO expert offers too, and a service sold as AEO or GEO is only as good as how closely it matches the way Search actually works. On ranking first, Google itself is blunt: no one can guarantee it.

An absolute guarantee clashes with the nature of the work. The SEO you trust with your project works on the site, the content, and the structure, and works to make sure external sources represent the brand consistently, through digital PR and fixing information where possible; the final call belongs to the system and happens inside a market where competitors and demand keep shifting too. A professional can promise you a method: an explicit hypothesis, actions that match that hypothesis, defined signals to test it, and the willingness to change course when the data proves it wrong.

When you ask yourself who should handle SEO on your project, the answer is rarely a single role. Some projects have a technical bottleneck, others an editorial one, others still something tied to reputation, local presence, or AI answers. The SEO specialist’s job is to spot which condition is limiting the result and to coordinate the skills needed to move it.

When results show up

Some changes can show effects within a few hours, others take several months.

When you read a result, the gap between the change you made and the signal you see changes what that result means. An effect that shows up too soon after a change may depend on something else; one that takes longer may simply not have kicked in yet. SEO timelines matter in this verification just as much as ranking, traffic, and conversions, and they make any attribution made the day after a change premature.