Search engine optimization (SEO) is the work of making a website easy for search engines to find, understand and show for relevant queries. Google says the same practices apply to AI Overviews and AI Mode, and that a page needs no extra requirements beyond being indexed and eligible for a snippet. Pew Research Center found that in March 2025, users clicked a result link in 8% of visits that showed an AI summary, against 15% without one.
This guide gives a site owner or marketer a working model of SEO. By the end, the reader can run a first technical check, choose which pages to build or fix, and read search data well enough to tell whether the work is paying off. It relies on Google's own documentation, Pew Research Center, web.dev, Ahrefs and Muck Rack, and it uses EPR's own Search Console data for one worked example.
What do search engines do before a page can rank?
A search engine finds a page, stores it, and then decides when to show it, and a page can fail at any of those three stages. Google's documentation on AI features and your website states that a page must be indexed and eligible to be shown in Google Search with a snippet to appear as a supporting link in AI Overviews or AI Mode.
That sentence describes the sequence every SEO task serves. The crawler has to reach the page, the index has to hold it, and the ranking systems have to judge it a good match for a query. Work done at a later stage cannot rescue a failure at an earlier one, which is why a well-written page that is blocked from indexing earns nothing.
| Stage | What happens | What commonly goes wrong | Where to check |
| Crawl | Google's crawler requests the page's URL | Blocked by robots rules, broken links, no internal links pointing to the page | Search Console URL Inspection tool |
| Index | Google stores and interprets the page | A noindex tag, duplicate pages, thin content | Search Console Pages report |
| Rank and show | Google selects results for a specific query | The page does not match what the searcher wants | Search Console Performance report |
Does SEO still matter when AI answers the question?
SEO still matters because Google's AI features choose their supporting links from the same indexed, snippet-eligible pages that ordinary results use. Google's guidance says the existing SEO fundamentals remain worthwhile for AI Overviews and AI Mode, and that there are no additional technical requirements to appear in them.
What has changed is the return on a ranking. Pew Research Center tracked 68,879 Google searches by 900 U.S. adults in March 2025. Users clicked a traditional result link in 8% of visits that showed an AI summary and 15% of visits that did not, and about one in five searches in the sample produced an AI summary, according to the Pew Research Center short read. A page can hold the same position and receive fewer visits when an AI summary sits above it.
The practical conclusion is to treat a ranking as a requirement and not as the finish line. A site still needs to be findable, and it also needs a reason for a searcher to click or for an AI answer to name it. For a comparison of the two disciplines, see GEO vs SEO, and for the vocabulary around answer engines, see the difference between AEO, GEO and SEO.
What are the three parts of SEO work?
SEO work divides into technical health, content and authority, and each part answers a different question about why a page is not performing. Sorting problems into these three parts keeps a team from rewriting content when the real issue is that the page is not indexed.
| Part | Question it answers | First checks | Main report |
| Technical health | Can Google reach, index and display the page? | Indexing status, internal links, page speed, snippet settings | Search Console Pages report |
| Content | Does the page answer the query better than the pages above it? | Search intent, depth, evidence, clear headings, current facts | Search Console Performance report |
| Authority | Do other credible sites link to or mention the brand? | Linking sites, brand mentions, outlet quality | Search Console Links report |
A small site with limited authority usually gets the fastest gain from the first two parts. Authority takes longer to build because it depends on other people choosing to cite the site, which later sections cover. A professional services firm can see how the three parts combine in why professional services firms need SEO, and a business serving one area can follow local SEO tips.
Which technical checks come first?
The first technical checks confirm that Google can index each important page, that internal links lead to it, and that the page loads and responds within Google's published thresholds. These checks come before any content work because they decide whether the content can be seen at all.
- Confirm in the Search Console Pages report that each important page is indexed and is not excluded by a noindex tag.
- Check that every important page has internal links from other pages on the site. Google's guidance lists making content findable through internal links as a fundamental practice.
- Confirm that the page can show a snippet, because Google requires snippet eligibility for a page to appear as a supporting link in AI features.
- Measure the three Core Web Vitals and compare them with the published thresholds.
- Read the page on a phone, since Google reports Core Web Vitals separately for mobile and desktop.
The Core Web Vitals thresholds are published on web.dev. Largest Contentful Paint, which measures loading, should occur within 2.5 seconds. Interaction to Next Paint, which measures responsiveness, should be 200 milliseconds or less. Cumulative Layout Shift, which measures visual stability, should be 0.1 or less. Each is judged at the 75th percentile of page loads, which means three quarters of real visits should meet the target.
The web.dev guidance describes INP as replacing First Input Delay as a Core Web Vital in 2024. A site that still reports First Input Delay is using an out-of-date measure. The full technical review, including the order to audit in, is covered in the SEO audit playbook.
How do you decide what to write or fix?
The clearest decision rule is to start with pages that already earn impressions but few clicks, because those pages have proven demand and a specific weakness. EPR's own Search Console data shows the pattern. Between July 16 and October 13, 2026, two of its SEO guides sat at average positions of 25.0 and 21.6, and together they earned 5,001 impressions and 1 click.
Those two pages were visible to Google and matched real searches, yet almost nobody chose them. The likely causes are a weak match with what searchers wanted and a ranking position below where most clicks happen. Neither cause is solved by publishing a new page. Both are solved by improving the page that already has the impressions.
| What Search Console shows | Likely cause | Fix to try first |
| High impressions, near-zero clicks, position 11 or lower | The page covers the topic but does not answer the specific query well | Rewrite the opening to answer the query directly, add evidence, and cover the questions the top results cover |
| Position 1 to 10, clicks below similar pages | The title or meta description does not promise an answer | Rewrite the title and description to state the payoff |
| Almost no impressions | The page is not indexed, or the query has little demand | Check indexing, then check whether people search the topic at all |
| Impressions fall after a model or ranking change | A competing page, an AI summary or an update changed the results | Compare the current results page with the previous one before editing |
Search intent decides which fix applies. A searcher typing a definition wants a short answer and a plain explanation. A searcher comparing options wants a table and a clear verdict. Open the results page for the target query, read the top few pages, and match the form they use before adding anything new. When two of a site's own pages compete for the same query, merge them instead of keeping both, because two pages chasing one term split the impressions between them.
How should a page be written to rank and get cited?
A page should open with a direct answer to the query, support it with named evidence, and use headings and tables that match how searchers ask the question. Google's guidance on AI features names creating helpful, reliable, people-first content as the key best practice, and it adds no AI-specific writing requirement.
Four habits follow from that guidance. First, answer the main question in the first two or three sentences, because a searcher and an answer engine both read the top of the page first. Second, phrase headings the way a person would ask the question, and put a self-contained answer in the first sentence under each one. Third, attach a named source and a date to every statistic, so a reader can check it. Fourth, use a table when the content is a comparison, a process or a set of criteria.
The form of the query changes the best form of the page. Muck Rack's July 2025 report found that advice-seeking and opinion-based prompts trigger more varied citations, while encyclopedic questions tend to fall back on older, static material. A page on a settled definition competes with reference sources, so it needs depth or a clear original angle. A page on a recent question can win by being current and specific.
What is semantic SEO, and does it change the work?
Semantic SEO is the practice of writing a page around a topic and the entities in it, such as people, products, organizations and places, instead of repeating one keyword phrase. It changes the writing habit more than the technical setup, because a search engine or an AI system has to work out what a page is about from the words on it.
Three habits put this into practice. Name each entity in full on first mention and use the same name afterward, so the page does not rely on pronouns to carry meaning. Define a term in one plain sentence where it first appears, so the page can be understood on its own. Cover the closely related questions a searcher would ask next, so the page resolves the topic instead of touching it. Structured data, which is code that states facts about a page in a standard format, is a separate and optional step. Google's guidance on AI features says no additional requirements or special optimizations are necessary to appear in AI Overviews or AI Mode, so markup is a way to state facts clearly and not a condition for appearing.
What mistakes waste SEO effort?
The most common waste is splitting one topic across several pages, because each page earns a fraction of the impressions that one strong page could hold. EPR's Search Console data from July 16 to October 13, 2026 shows the effect in its own library. Four pages on how AI changes SEO earned 313, 324, 363 and 242 impressions, at average positions between 14.8 and 44.4, and none of them averaged a first-page position.
Other common mistakes are easier to spot with the reports named above. Publishing new pages before fixing indexing leaves the new pages invisible. Rewriting a page that already ranks in the top three without a clear reason can lose the position it earned. Chasing links through purchase or exchange breaks Google's spam policies. Reading a fall in clicks as a content failure, without checking whether impressions held, can send a team to fix a page that is working.
How do links and mentions build authority?
Links and mentions build authority because they show that other sites find a page worth citing, and the evidence for mentions has grown stronger than the evidence for raw link counts. Ahrefs studied 75,000 brands and found that branded web mentions correlated with AI visibility at 0.66 to 0.71 across ChatGPT, Google AI Mode and Google AI Overviews, with YouTube mentions highest at about 0.737. The Ahrefs correlation study notes these are correlations, not proof of cause.
Citation data from the public relations side points the same direction. Muck Rack's May 2026 edition of its What Is AI Reading? report found that earned media made up 84% of citations across ChatGPT, Claude and Gemini, and paid or advertorial content made up 0.3%. Muck Rack defines earned media broadly, and its figure includes journalism, academic research, government sources, encyclopedic sites and third-party corporate content. The full method for earning that coverage is in the PR and SEO plan.
Google's spam policies set the limit on how to pursue links. They list buying or selling links for ranking purposes and excessive link exchanges as link spam, and they say paid links are acceptable when marked with a rel="nofollow" or rel="sponsored" attribute. The safe route is to publish pages with original data or clear explanations that other sites choose to cite. The methods for that are covered in link building and digital PR.
How do you measure SEO now?
SEO is measured with Search Console for rankings and clicks, plus a separate check of whether AI answers name the brand. Google's documentation says that sites appearing in AI features are included in the overall search traffic reported in Search Console, so AI Overview and AI Mode appearances are not reported as a separate channel there.
That means a drop in clicks can come from an AI summary absorbing the click while the page keeps its position, and the Pew data above shows the size of that effect in the March 2025 sample. Compare impressions and position against clicks before drawing a conclusion. If impressions hold and clicks fall, the page is still being shown and fewer people are choosing it.
| Measure | Source | What it tells you |
| Impressions and average position | Search Console Performance report | Whether Google shows the page and where |
| Clicks and click-through rate | Search Console Performance report | Whether searchers choose the page |
| Indexed pages and exclusions | Search Console Pages report | Whether Google can hold the content |
| Core Web Vitals pass rate | Search Console and PageSpeed Insights | Whether loading, response and stability meet the thresholds |
| Sites linking to key pages | Search Console Links report | Whether other sites cite the content |
| Share of AI answers naming the brand | Repeated prompt checks or a monitoring tool | Whether AI answers include the brand for buyer questions |
For the AI answer check, choose 25 to 50 questions that real buyers ask, run them on a regular schedule across ChatGPT, Gemini, Claude and Google AI Mode, and record whether the brand is named and which pages are cited. Muck Rack notes that the behaviors in its reports may shift as models are updated or retrained, so record the date of each run and repeat prompts instead of relying on a single answer. A team that needs to explain these numbers to senior leaders can use the format in how to explain SEO, GEO and AI visibility to executives.
What should a site do in the first 90 days?
A site can follow three 30-day phases that move from technical repair to content improvement to measurement. The plan needs only Search Console and a spreadsheet, and it makes each phase depend on the one before it.
| Days | Actions | Output |
| 1 to 30 | Check indexing for every important page. Fix blocked, excluded or orphaned pages. Measure Core Web Vitals on the main templates. Confirm snippets are allowed. | A list of fixed technical issues and a baseline report |
| 31 to 60 | Export pages with high impressions and low clicks. Rewrite openings and titles to answer the query. Merge pages that compete for the same term. Add internal links to improved pages. | A set of updated pages and a record of what changed |
| 61 to 90 | Compare impressions, position and clicks with the baseline. Run the AI answer prompt set. Publish one page with original data that other sites could cite. | A before-and-after report and a plan for the next quarter |
The order is deliberate. Fixing indexing first means later content work can be seen. Improving existing pages second uses the impressions a site already has before spending effort on new pages. Measuring last separates changes that moved the numbers from changes that only felt productive.