An SEO strategy for 2027 has one job with two audiences: rank in Google and Bing, and be the page that AI Overviews, ChatGPT, Gemini, Perplexity and Copilot choose to quote. The foundation for both is the same, and most of the AI “hacks” sold on top of it have no evidence behind them.
The short answer
- Classic SEO is not obsolete. Google’s AI Overviews and AI Mode run on Google’s index, and in one large ChatGPT study the first result its own search returned was cited about four times as often as the tenth.
- Clicks can fall even when visibility holds, so I measure search visits, AI mentions and qualified enquiries separately. The size of that shift depends on the query and audience.
- AI search can expand one question into related searches. Cover the buyer’s connected questions clearly, without manufacturing a separate page for every wording.
- Brand mentions, YouTube and reviews track AI visibility more closely than backlink counts do.
- llms.txt files, schema added to win citations, year-in-title tweaks and hidden “AI instructions” have no reliable effect, and some breach Google’s spam policies.
What is an SEO strategy in 2027?
An SEO strategy in 2027 is a plan to make a business the most retrievable, most quotable and most-mentioned source for a defined set of buyer questions, across Google, Bing and AI assistants. It covers the technical foundation, the content that answers those questions, the reputation built off-site, and how results are measured.
I have built on the web since 2005. The SEO advice I published in 2015 was mostly about keywords, meta tags and social signals. This page replaces a dozen of those old posts. Here is what the strategy looks like for 2027, in four layers:
- Foundation. A fast, crawlable, secure site that search engines and AI crawlers can read and are allowed to read.
- Answers. One strong page per buyer question cluster, written answer-first, with headings that match how people ask.
- Reputation. Reviews, mentions, video and earned links that make engines recognise your name.
- Measurement. Search Console, Bing’s AI reports and the leads that say how they found you.
For a small business, search remains useful because a relevant query reveals a need you may be able to meet. Some people now read an answer without visiting. In a February 2026 US survey, Pew Research found that 60% of adults said they ever read AI summaries at the top of search results. That measures reported use, not lost clicks or Australian buying behaviour.
Does SEO still matter when AI answers the question?
SEO still matters when an AI answer uses web search. Google’s AI features rely on its Search index and core ranking systems, while other products use their own retrieval arrangements. Strong Google rankings are useful visibility, but are not the same thing as a place in ChatGPT’s retrieved results or a citation. Some assistant responses do not search the web at all.
Google’s current AI search guide keeps normal SEO at the centre. Eligibility needs an indexed page that can show a snippet, plus inclusion in Search generative AI features in Search Console. Include is the default, but a property can inherit an exclusion from its parent. Check the effective setting; it is separate from training controls such as Google-Extended. Eligibility still does not guarantee display.
- In AirOps’ April 2026 study (16,851 queries run three times each through the ChatGPT interface, 50,553 answers), the first result ChatGPT’s own web search returned was cited 58.4% of the time, against 14.2% for the tenth. That retrieval order is ChatGPT’s search ranking, not a Google position, and other engines may behave differently.
- Across 115,254 sites, Ahrefs measured Google sending 23.96% of all website visits in September 2026. All AI assistants combined sent 0.43%.
- The click picture varies. SparkToro’s Similarweb-panel analysis put US zero-click Google searches at 68% for January to April 2026. A separate Ahrefs study associated AI Overviews with 58% lower average desktop CTR for the top result, comparing December 2023 and December 2025 keyword samples. Neither figure predicts the loss for an Australian business.
- Ranking is not a citation guarantee. In Ahrefs’ March 2026 analysis of 863,000 keyword SERPs and four million AI Overview URLs, 38% of cited URLs also ranked in the top ten for the original query. Fan-out and differences in parsing complicate comparisons with older studies.
I would keep classic search and AI visibility in one plan, while reporting them separately. Abandoning pages that already help buyers find you to chase an untested GEO tactic is a poor trade. So is assuming that a Google position proves which sources an assistant will choose.
How do AI Overviews and ChatGPT choose which pages to cite?
Query fan-out is one way AI search gathers evidence: it runs related searches, retrieves pages and uses relevant information to build an answer. The retrieved page, the passage used and the citation shown are distinct steps. A clear heading and a direct answer help readers and can support retrieval, but no heading format makes every engine cite a page.
Google documents fan-out in its own guidance, and the Gemini API documentation says the model generates one or more search queries before it answers. In AirOps’ ChatGPT data, 88.6% of queries generated exactly two sub-queries and only 8.8% generated none. What that means for your pages:
- Answer first. In Surfer’s sample, 38% of AI Overview citations came from a page’s first 100 words. That is an observation about where cited text sat, not proof that moving text up earns a citation.
- Match the reader’s question. AirOps observed more citations where headings matched the query closely. That is an association in its ChatGPT sample, not a reason to copy exact phrases mechanically.
- Stay relevant. AirOps found focused pages performed well in its sample. Do not turn that into a percentage-of-subtopics target: a reader still needs the answer to be complete.
- Name your constraints. State the country, company size, platform and budget band you serve. Fan-out sub-queries carry those constraints. This is unmeasured, but low risk.
- Do not build a page per sub-query. Google’s spam policies treat mass pages made without adding value as scaled content abuse.
Which Google ranking factors still matter in 2027?
The Google ranking factors that still matter going into 2027 are the ones tied to usefulness and access: relevant content that answers the query, a crawlable and indexable site, clear titles and headings, mobile-friendly pages with good page experience, and links and mentions from sources people trust. Tricks built on keyword counts and link volume have aged badly.
I went back through the ranking-factor and SEO-tips posts I wrote between 2015 and 2017 and checked each piece of advice against current guidance.
| Old advice (2015 to 2017) | Status in 2026 | What to do instead |
|---|---|---|
| Put the keyword in the title tag and H1 | Still useful | Write a title that matches how people ask, keyword early, short enough not to be cut off |
| Add keywords to the meta keywords tag and repeat them through the copy | Obsolete. Google ignores the meta keywords tag | Cover the topic and its sub-questions in natural language |
| Write at least 400 to 600 words, longer is better | Myth. Google has no preferred length | Answer the question completely at the length it needs; focus and breadth depend on the reader |
| Exchange or buy backlinks | Against Google’s spam policies | Earn links and mentions with original data, tools and expertise |
| Submit an XML sitemap | Still useful | Generate it automatically and submit it in Search Console |
| Integrate Google Webmaster Tools | Renamed Google Search Console in 2015 | Use Search Console, including its generative AI performance report |
| Set up Google Authorship | Dropped by Google in 2014 | Build a real author entity: a bio, consistent profiles, Person schema for clarity |
| Grow social signals, including Google+ | Google+ closed in 2019. Shares are distribution, not a ranking lever | Use social and video to build familiarity with your name |
| Make the site mobile-friendly | Now the default. Google indexes the mobile version first | Design and test mobile-first |
| Publish fresh content often | Partly true. Real updates help; fake date changes are a warning sign | Update pages when the facts change |
On freshness, Ahrefs’ July 2025 study found pages cited by four assistants were younger on average than organic results. Its 25.7% difference measured age since publication, not a 25.7% uplift from updating a page; AI Overviews behaved differently. Google warns against changing dates without substantial edits. Update facts when they change and keep the date honest.
What SEO basics should your web developer know?
Your web developer should know the technical SEO basics that decide whether search engines and AI crawlers can find, read and trust a site: fast pages, clean semantic HTML, crawlable links, correct redirects and canonicals, an XML sitemap, and server settings that do not silently block search or AI bots.
This is the layer I care most about, because a business owns it outright. At WP Creative, the Diagnose phase of our WPO Framework includes a technical SEO review and a Core Web Vitals analysis before anyone talks about content. We are not a full-service SEO agency. We build and run the foundation that SEO and AI visibility sit on, through our WordPress development work. The checklist my team works from:
- Speed and Core Web Vitals. Aim for LCP of 2.5 seconds or less, INP of 200 milliseconds or less and CLS of 0.1 or less at the 75th percentile. Measure mobile and desktop separately. A good result supports page experience; it does not guarantee a ranking.
- Semantic HTML. One H1, headings in order, real lists and tables. Google’s guide to agent-friendly websites says AI agents read screenshots, the DOM and the accessibility tree.
- Crawlable links. Use <a> elements with valid href attributes. JavaScript may insert those links, but a click handler on another element is not a reliable substitute.
- Redirects. Use permanent server redirects such as 301 or 308 for a permanent move. Test that old URLs reach the closest replacement without loops or chains. A redirect communicates the move; it is not a guarantee that rankings carry over unchanged.
- Canonicals. One canonical URL per piece of content, so duplicates do not split the signals.
- Sitemap. Generated automatically, listing only indexable URLs. The old “100 to 150 links per sitemap” rule is wrong: Google accepts up to 50,000 URLs per sitemap file.
- Bot access. Check robots.txt and CDN rules together. Cloudflare’s July 2025 change applied AI-crawler blocking by default to newly added domains. OpenAI distinguishes OAI-SearchBot from its training crawler: blocking the search bot excludes ChatGPT search answers, although navigational links can remain. Allowing it does not promise a citation.
- Structured data. Keep supported markup accurate and consistent with the visible page. It can support eligible search features, but there is no special schema required for Google’s AI answers.
Design and build choices decide most of this list. I cover them in how web design affects SEO, and the platform-specific steps in the WordPress SEO settings I change on every site.
Do brand mentions matter more than backlinks for AI search?
Brand mentions track AI search visibility more closely than backlink counts in the best correlation data available, which is not the same as proving they cause it. Ahrefs found mentions and YouTube presence tracked AI visibility far more closely than backlinks across 75,000 brands. Backlinks still matter for classic rankings, which in turn feed AI retrieval, so treat the two as complementary.
- In the Ahrefs study, brand mentions correlated with AI visibility at 0.66 to 0.71 and YouTube mentions at 0.71 to 0.74. Backlinks sat at 0.2 to 0.3. That is correlation, not proof.
- A useful fresh piece of research is Chen and Koudas’ October 2026 preprint. Across two commercial markets, familiar brands appeared in rewritten searches more often. The authors distinguish brand mentions from page citations and describe changing a model’s prior as a hypothesis, not a tested marketing intervention. That supports investigating which sources an engine retrieves; it does not prove that manufacturing mentions will change its recommendations.
- For an Australian business, mentions come from a complete Google Business Profile with real reviews, directories where buyers in your category look, YouTube videos with transcripts, podcasts and talks, and the third-party lists AI engines already cite. Being genuinely helpful in digital marketing forums and communities earns mentions too.
- Bought or fake mentions do not hold up. Google’s AI guidance says inauthentic mentions are less helpful than they look.
Links still earn their place when they are earned. I keep my notes on what still works in link building separate, because it is its own discipline.
What changed in SEO between 2015 and 2026?
SEO changed from a game of keywords and links to a question of usefulness, access and reputation between 2015 and 2026. Mobile became the default, Google learned to read language rather than match strings, answers moved onto the results page, and from 2024 AI-generated overviews began answering many questions before any click.
| Year | What changed | What it means for your strategy now |
|---|---|---|
| 2015 | Google’s mobile-friendly update. My 2015 trends post called mobile the top trend | Mobile is now the default version Google indexes |
| 2016 to 2019 | Voice search hype and featured snippets (“position zero”) | The lasting lesson: conversational headings and answer-first passages |
| 2019 | BERT language understanding comes to Google Search | Write naturally. Keyword-matching tricks stopped paying |
| 2022 | Google’s helpful content system launches, later incorporated into core ranking systems in 2024 | Review usefulness page by page and across the site; thin archives are not an automatic sitewide penalty |
| 2024 | AI Overviews launch in the US | Being quoted matters as much as being clicked |
| 2025 | Google AI Mode launches in Australia | Australian buyers get long, conversational AI answers inside Google |
| 2026 | Spam policies cover manipulating generative AI responses. Search Console adds generative AI reports | GEO hacks carry real risk. AI impressions become measurable |
For the dated update record, see my history of Google algorithm updates.
Which GEO tactics should you ignore?
The GEO (generative engine optimisation) tactics to ignore are the ones with no measured effect or a spam risk: llms.txt as a ranking lever, schema added to win citations, year-in-title tweaks, date bumps without real edits, one page per fan-out query, mass AI content and hidden instructions for AI crawlers. None has reliable evidence behind it.
| Tactic | Verdict | Evidence |
|---|---|---|
| Add an llms.txt file to rank in AI answers | No Google ranking benefit | Google ignores it. Ahrefs sampled 137,210 domains; 28% had a valid file and 97% of those files received no requests in May 2026 (study) |
| Add schema to win AI citations | No reliable citation uplift established | Ahrefs matched 1,885 pages adding JSON-LD against controls. Its small AI Overview decline does not establish that schema causes harm (May 2026 study) |
| FAQ schema for extra search space | Myth | Google no longer shows FAQ rich results (Search Central updates) |
| Put the year in the title | Inconclusive | An untouched control page rose as much as the edited pages (Otterly) |
| Hidden “AI instructions” or white text | Spam | Google’s spam policies now cover attempts to manipulate generative AI responses. Engines did not obey the hidden text (Otterly) |
| One page per fan-out query, or AI content at scale | Spam risk | Scaled content abuse under Google’s spam policies |
| Track your “AI rank” from one prompt | Too little evidence | SparkToro and Gumshoe’s January 2026 research found recommendation lists varied across repeated runs. Repeated presence was more useful than an individual list’s order |
The honest version of GEO is just good SEO plus reputation. Be retrievable, be quotable, be known.
How much weight should you put on these studies?
Treat every number in this article as a signal from one study, with its own engine, sample and method, not a law of search. Here is what the main ones can and can’t tell you.
| Study | What it measured | What it can’t tell you |
|---|---|---|
| AirOps fan-out study (April 2026) | 16,851 queries run three times each through the ChatGPT interface: sub-queries, retrieval order and which pages were cited | How Google, Gemini, Perplexity or Copilot behave. Retrieval order is ChatGPT’s own search ranking, not a Google position |
| Ahrefs AI vs search traffic (September 2026) | Share of visits to 115,254 sites from Google and from AI assistants | Whether those visits converted, or how your sector compares |
| Ahrefs brand visibility correlations | How mentions, YouTube presence and backlinks correlate with AI visibility across 75,000 brands | Cause and effect. Bigger brands have more of everything |
| Surfer first-100-words analysis | Where in a page the cited text sat, for AI Overview citations in its sample | Whether rewriting your opening earns a citation |
| Pew Research Center (June 2026) | February 2026 survey: 60% of US adults said they ever read AI summaries at the top of search results | How Australian buyers behave, how often summaries are read, or how many clicks they replace |
How do you measure an SEO strategy on your own site?
Measure your site from a dated baseline, because another business’s study cannot show what changed for yours. In the WPO framework, I use data to separate the questions: actual search clicks, estimated visibility, AI mentions and qualified enquiries. Each has its own method and denominator. Record the date, market and collection window, then compare like with like. A rise in citations is not evidence of more leads, and a third-party traffic estimate is not a count of visits.
- Baseline first. Export Search Console clicks and impressions for your question pages, with the date, before you change anything.
- AI visibility as a trend. Use Search Console’s generative AI report and Bing’s AI Performance report, plus a fixed set of buyer prompts run several times each month. One run tells you nothing.
- Leads by source. A “how did you find us?” field on the enquiry form, read every month.
- Compare like with like. Record the change date, then review after about 30 days as an early checkpoint. Keep the same market, device and reporting window, allow for conversion lag, and compare with seasonality where possible. A redesign, host move or Google update in the same period prevents a clean claim that one content edit caused the result.
What to do next
- List the 20 to 30 questions buyers ask before they contact you, and give each one a single owning page on your site.
- Fix the foundation first: speed, crawlable links, redirects, sitemap, and bot access in both robots.txt and your CDN.
- Rewrite your key pages answer-first: a question-shaped heading, then a short answer that stands on its own.
- Merge thin, overlapping old posts into one strong page each, with 301 redirects from every old URL.
- Build familiarity off-site every month: reviews, Google Business Profile, YouTube, talks, and the lists AI engines already cite.
- Measure with Search Console’s generative AI report, Bing’s AI Performance report and a “how did you find us?” field on your enquiry form.
- Want a second opinion on your strategy or your site’s technical foundation? Book a free call with me and my team.
Frequently asked questions
Is SEO still worth it for a small business?
SEO is still worth it for a small business, because people who search for what you sell already have the problem you solve. The difference now is that fewer of them click. A small business should fix its technical basics, answer its buyers’ real questions clearly, and collect reviews, so it shows up both in the results and in the AI answer above them.
How long does an SEO strategy take to work?
There is no fixed SEO deadline. A technical change may be reflected after recrawling, while content and reputation can take longer and may not produce a measurable gain. Use a 30-day review as a diagnostic checkpoint, then judge commercial results over a period that fits your sales cycle. Record other changes and repeat AI prompt checks rather than treating one run as a trend.
Do I need an llms.txt file for AI search?
Google does not use llms.txt for Search visibility. In Ahrefs’ June 2026 study of 137,210 domains, 28% had valid files and 97% of those files had no requests during May. That does not prove no service reads them. My priority would be useful pages and reliable discovery; maintain a file if a specific tool you use benefits from it, not because someone promises it will make assistants cite you.
Is voice search still worth optimising for?
Voice search is worth optimising for only as part of a wider shift to conversational queries. The voice-search wave of the late 2010s taught one durable lesson: people ask full questions. AI prompts are longer than typed searches, so headings that mirror real questions and answers that stand alone serve voice assistants, AI Overviews and chat assistants at the same time.
Do you sell SEO services?
I do not sell SEO, GEO or AI optimisation services. My team builds and runs the technical foundation that search and AI visibility depend on: fast, secure, well-structured websites. For content and link campaigns, we work alongside specialist SEO partners. If you want to talk through your strategy or your site’s foundations, a free call with me and my team is the place to start.
