What Actually Works in SEO and AI Search: 17 Tactics Backed by 1,391 Measured Case Studies

Key takeaways

• Content refreshes have some of the best evidence in organic search. A controlled test on 40 pages found updating facts and adding missing sections lifted organic sessions by 55% (ContentPulse).
• FAQ content on category pages pointed the same way in all 15 samples the review found, with a controlled +9% from SearchPilot. But Google no longer shows FAQ rich results, so the value now sits in the content, not the snippet.
• Trust signals appear to move AI citations. Adding author bylines to blog posts was associated with a 113% rise in AI citations on 123 pages compared with a control group (Seer Interactive).
• Writing for AI answers means adding evidence, not keywords. In the GEO benchmark study, adding statistics and cited sources raised visibility, while keyword stuffing lowered it by 10% (Aggarwal et al., arXiv).
• Some popular "optimizations" backfire. Rewriting title tags around higher-volume keyword variants lost traffic in controlled tests from Moz (−20%) and SearchPilot (−11%).
• Most SEO case studies can't prove causation. 51% of the tactics in the review rest on at least one study that changed several things at once, and published results skew heavily positive.

These takeaways come from The SEO Meta Study by Krešimir Ćorluka of Ovisy (September 2026), which screened 9,249 SEO, AEO and GEO documents and kept only the 1,391 that measured a real change. The result: 182 tactics graded on the evidence behind them.

Why this study is different from a "ranking factors" list

A ranking factor list tells you what correlates with top positions. This review does something narrower and more useful for practitioners: it only counts documents where someone changed something and measured what happened. Three design choices are worth understanding, because they also make you a better reader of any case study.

Evidence and actionability are scored separately

Each tactic gets two scores. E (evidence) reflects how strong the proof is: study design, number of independent samples, sample size and how consistently results pointed the same way. A (actionability) reflects how cheap, safe and reversible the change is. The review deliberately never merges them into one verdict, since a well-proven tactic can still be a poor use of your week. For its shortlist it requires E of at least 60 and then ranks by 0.35 × E + 0.65 × A, on the logic that among adequately supported tactics, the useful question is which one pays off first.

A controlled test counts for far more than a before/after chart

Study designs are weighted by how much of the change they can attribute to the intervention. A controlled split test carries 25 points; a before/after measurement with no control group carries 3. That gap reflects reality: a before/after chart can't separate your change from seasonality, a core update, or the three other things your team shipped that month.

Agreement is shrunk so tiny samples don't look certain

If one study reports a positive result, raw agreement is 100%. The review applies a simple correction, (k + 1) / (n + 2), so one out of one becomes 0.67 and nine out of eleven becomes 0.77. It's a useful mental habit: a single impressive case study is a hypothesis, not a law.

The honest caveats the authors publish themselves

Ćorluka is unusually open about the limits of the evidence base, and these points are worth knowing before you read any SEO case study:

• Publication bias is large. In the author-verified reference sample, 64% of reported outcomes were positive, 16% mixed, 12% null and just 8% negative. SEO practitioners write up the wins, so every agreement figure should be read as an upper bound.
• Half of all tactics involve bundled changes. 92 of the 182 entries (51%) rest on at least one source that changed something else at the same time. For those, trust the direction more than the size.
• Almost nobody reports uncertainty. Only 4 of 25 usable studies in the reference sample reported a confidence interval, which is why the review reports medians and ranges instead of running a statistical meta-analysis.
• Controlled tests are rare and concentrated. A single publisher accounts for roughly one source in eleven and most of the controlled split tests in the corpus. You'll see SearchPilot's name repeatedly below.
• Much of the evidence is recent. Around two thirds of the corpus was published in the last two years, and AI-search findings are given a shorter shelf life than classic on-page findings.

The 17 tactics on the shortlist

Of the 61 tactics that met the evidence threshold, these 17 ranked highest on the combined score (no more than three per metric family). The effect shown is the headline figure the review printed for that entry; "bundled" means the source changed several things at once.

* Reported by the source as a percentage change in ranking. Table source: The SEO Meta Study, Krešimir Ćorluka, Ovisy.

Below, the tactics are grouped by the job they do, with the original studies behind each one.

1. Refresh what you already have

If you only act on one section of this article, make it this one. Updating existing content is cheap, reversible and among the better-evidenced tactics in the review.

Update facts and expand thin sections. The strongest source here is a controlled test on 40 pages in which ContentPulse updated outdated facts and expanded sections on a group of posts, compared with untouched control posts. Organic sessions rose 55%. Supporting evidence comes from Authority Hacker, where a single updated page climbed 43 positions, and Backlinko.

Refresh the whole page, including the date. A second entry covers fuller refreshes that update all of these together:

• Title tag
• Meta description
• Headings
• Body copy
• Internal links
• Publication timestamp

The reported results:

• Inflow: +57% impressions on one URL
• Ahrefs: +36% on a republished page
• Backlinko: +260.7% on a relaunched page
• Moz: +5.89% across 41 re-optimized pages

The Moz number is the most useful one. Plan for the Moz outcome and treat the Backlinko outcome as the upside.

What this means in practice: changing the date alone isn't what these studies tested. The changes that were measured involved genuinely new information. Build a refresh queue from pages that have slipped in impressions, have outdated figures, or miss subtopics competitors now cover.

How PostKing helps: PostKing pulls your Search Console and GA4 data into a ranked "do this next" worklist, so decaying pages surface on their own. Its content gap analysis shows the topics competitors cover that you don't, which is exactly the missing-sections list a refresh needs.

2. Answer the questions people actually ask

FAQ content on category pages. This entry has the most consistent direction in the entire review: 15 of 15 samples pointed the same way, with a median of +112% organic sessions across six samples and a range from +9% to +350%. Only the low end comes from a controlled test. SearchPilot added a question and answer about top brands to category pages, plus matching FAQ schema, and measured +9% organic sessions against a control group. The large figures come from uncontrolled cases, such as a single-page CTR test by seoClarity (+350%) and a 5,000-page AI Overviews project by Conversion Digital (+41% citation rate).

An important update: the review notes that Google stopped showing FAQ rich results in May 2026, citing Search Engine Land. Much of the older FAQ evidence captured a SERP feature that no longer exists. The on-page question-and-answer content can still help users and answer engines; just don't expect the expandable snippet.

FAQs, tables and structure on product pages. HubSpot reported a 56% increase in AI citations after structural changes of the kind this entry covers: FAQ sections, direct headlines, feature tables and structured markup. Two controlled tests from SearchPilot back the direction for organic search: +20% sessions in one test and +24.5% CTR from enriching page content.

3. Show who wrote it and why they're credible

Author bylines. Seer Interactive added bylines to blog posts, compared them against a control set, and recorded a 113% increase in AI citations across 123 pages. A detail worth noticing: the bylines belonged to staff members without much public profile, so the effect wasn't just borrowed celebrity. A SearchPilot controlled test on authorship pointed in the same positive direction, and all seven samples in this entry agreed.

Human editing of AI drafts. In a controlled experiment across 10 domains, Reboot Online found human-written content ranked about 2.2 positions better than AI-written content. A later, much smaller Reboot test saw AI citations rise 67% after an editor rewrote AI text with credentials, verified sources and direct answers. On the commercial side, an academic controlled study published via Neliti found human-written product descriptions converted 88% better than AI copy over three months.

How PostKing helps: every article in PostKing can carry a named author from your team, so bylines are built in rather than bolted on. For AI-assisted drafts, PostKing writes in a voice profile trained on your own content, runs a humanize pass that strips stock AI phrasing, and can score a draft for how generic it reads before a human editor signs off.

Cite your sources. Adding citations to authoritative sources was one of the best-performing methods in the GEO benchmark (see section 4), and Reboot Online's long-running outgoing link experiment found that pages linking out to authoritative sites performed better than those that didn't.

How PostKing helps: PostKing articles include authority links to credible external sources, placed inline in the prose or as a reference list at the end, and can be grounded in the facts stored in your brand knowledge base.

4. Write for AI answers without writing for robots

Generative engine optimization has the thinnest live-site evidence and the most academic evidence, which is exactly why it's worth separating the two.

The anchor study is GEO: Generative Engine Optimization by Pranjal Aggarwal and colleagues, which tested content rewrites against 1,000 queries on a simulated generative engine. Relevant results reported in the review:

• Adding quantitative statistics raised citation visibility by 37%.
• Adding quotations from credible sources raised impressions between 22% and 41%.
• Adding citations to authoritative sources produced +31.4% (though across variants, results ranged from −30.3% to +115.1%).
• Keyword stuffing lowered visibility by 10%, the only method in this group the review marks as one to avoid.

Across two samples, full optimization-oriented rewrites produced a median +83.8% citation rate, and 12 of 12 samples in that entry pointed the same way. These results come from a simulated engine, which is why the live-site studies below matter.

The live-site evidence points the same way. In a controlled test on 30 matched URLs, Deepak Gupta replaced repeated target keywords with natural semantic variations and saw AI citations rise 180%. Seer Interactive reported a 344% jump in AI citations across 11 pages after simplifying copy and aligning it to intent (uncontrolled).

The practical pattern: answer the question early, back claims with specific numbers and named sources, vary your language naturally, and stop repeating the exact-match keyword.

How PostKing helps: PostKing tracks how often AI answer engines mention and cite your brand, alongside Bing's AI citation data, so you can see whether GEO rewrites are actually getting picked up. Its SEO and GEO pipeline writes articles built for both search rankings and LLM citations.

5. Add structured data, with realistic expectations

Schema markup appears in three shortlist entries, but nearly all of the large numbers come from uncontrolled case studies:

• JSON-LD on blog posts: Writer Sanctuary compared 25 marked-up pages against a control and reported +31.78% on rankings; Fueler reported +22% on 31 pages.
• Product and Article schema: Rank and Discover reported +40% organic sessions across 25 sites; a small controlled test by getAISO measured +30% on 2 pages.
• Product, Review and FAQ schema on SKU pages: SEO Melbourne reported +40% CTR; WordLift reported +48.8% on a legal services site.

Product and review markup still produce rich results; the FAQ part no longer does. Schema is a low-risk, reversible change, which is why it ranks well on actionability, but the controlled evidence for large gains is thin.

6. Fix the structure: internal links and consolidation

Link supporting articles to their pillar. Adding in-body anchor links from cluster articles to the parent pillar page was associated with a 94% ranking improvement across 18 sites in SEO Mafia Club's internal linking experiments, with supporting cases from Postion and Featured.

Internal-link CTAs convert. In a controlled test run with SEOTesting for BT, adding a brand-icon block linking product pages to brand category pages raised conversions by 80%, a win for both SEO and UX.

Merge thin, repetitive pages into hubs. Consolidating near-duplicate package pages into comprehensive procedure hubs, rewriting regional pages with distinct local details, and redirecting retired URLs was linked to +74% conversions by Digital Surfer. Inflow and Lumar reported similar directional gains.

How PostKing helps: PostKing's keyword clustering groups your topics into clusters with a pillar keyword, then plans and writes the supporting articles around it. It can also auto-assign call-to-action blocks that link published articles to the right landing page, and when you change a URL it rewrites the internal links pointing at the old one.

7. Local SEO: consistency beats cleverness

Keeping name, address, phone, hours and URL identical across directories and map services has a large sample behind it. Yext's work with Accor reported +38% organic clicks, and Yext's Sofology case reported a 33% gain. Updating the primary business category, menu details, local business schema and review prompts together was associated with an improvement of around 10 ranking positions.

Be careful with category changes, though. The review's "it depends" section shows moving to a more specific primary category gained 3.4 positions (Flento), while moving to a broader label lost around 30 positions. Always ask whether the new category is more specific than the current one.

What the evidence argues against

Swapping title tags to chase higher-volume keywords. This is the tactic the review explicitly marks "avoid". A Moz controlled test lost 20% of organic sessions, SearchPilot lost 11%, and a later SearchPilot analysis reported −27%. Only an uncontrolled 200-URL Moz case showed a gain (+9%). Specific, descriptive titles tend to beat generic high-volume phrasing.

Stuffing keywords for AI. As above, the GEO benchmark measured −10%.

Rewriting URLs around broad keywords. In the same Moz case study, changing URLs to target broader high-volume terms coincided with a 38% drop in organic sessions. Adding specific modifiers was a different story (Dave Ashworth). Changing established URLs carries redirect risk regardless.

When the answer is "it depends"

Some tactics split cleanly into two contexts, and averaging them would describe neither:

Worth testing, not yet worth trusting

The review's experimental section holds tactics with at most two independent samples. One example: pointing canonical tags on main product pages to a specific variant URL lifted organic sessions 22% in a SearchPilot controlled test, with a second test pointing the same way. Nobody outside that publisher has replicated it. Read tactics like this as hypotheses with a method attached.

How to test any of these on your own site

A result measured on someone else's site is evidence, not a forecast. The review structures every replication around six questions, which make a solid template for any SEO test:

1. When does it apply? Define the page type and the problem (for example, "blog posts older than 18 months with declining impressions").
2. Before you start: record a baseline and pick a matched control group of similar pages you won't touch.
3. What to change: make the change on the test group only. If you bundle several edits, accept that you're testing the package.
4. What to keep: freeze titles, internal links, templates and URLs on both groups during the test.
5. What to measure: choose one primary metric (organic sessions, clicks, impressions, AI citations) and a fixed window, commonly 28 to 60 days.
6. When to revert: write down, before launch, the result that would make you undo it. Without a revert rule, every outcome looks like confirmation.

Step 6 is the one most teams skip, and it's the one that protects you from publication bias inside your own organization.

How PostKing helps: steps 2 and 5 need clean before-and-after numbers. PostKing brings Search Console, GA4 and AI citation data into one place, so you can baseline your test pages and check both classic search and AI visibility without switching tools.

A red-flag checklist for reading SEO case studies

Use the review's logic whenever someone shares a dramatic chart:

• No control group? Seasonality and algorithm updates are not ruled out.
• Several changes at once? The number belongs to the whole package.
• One site, one page? It's an anecdote until someone replicates it.
• Percentage on a ranking? Check whether it's positions gained or a percentage of something else.
• Benchmark metric? Academic GEO results come from simulated engines, not live search.
• Only wins published? Assume the losses exist and weren't written up.

A prioritized plan for your next sprint

Based on evidence strength, cost and reversibility, here's a sensible order of operations:

1. Refresh decaying content with new facts, missing sections and updated internal links.
2. Add visible author bylines linked to real author pages on informational content.
3. Add concise Q&A sections to category and product pages, written for users, not for the retired FAQ snippet.
4. Rewrite key informational pages for answer engines: lead with the answer, add specific statistics and cited sources, replace repeated exact-match keywords with natural variation.
5. Link cluster articles to their pillar pages and add prominent internal-link CTAs where they help users.
6. Implement Product, Review and Article schema where it's missing.
7. Audit title tags that were rewritten around high-volume variants and test reverting them.
8. For local businesses, sync NAP data everywhere and choose the most specific accurate primary category.

Run each as a controlled test where your traffic allows it, and publish your null results. The field needs them.

Frequently asked questions

What is The SEO Meta Study?

It's a systematic review with narrative synthesis of SEO, AEO and GEO case studies, published by Krešimir Ćorluka and Ovisy in September 2026. It screened 9,249 documents, kept 1,391 that reported a measured outcome, and graded 182 tactics built from 552 primary sources. It is not a statistical meta-analysis, because most studies don't report the variance needed to pool results.

Which SEO tactic has the strongest evidence?

Among tactics with controlled tests, content refreshes, FAQ-style content on category pages, schema markup and author bylines score highest on evidence. The review ranks FAQ content on category pages first on its combined score, with all 15 samples pointing the same way.

Does FAQ schema still work in 2026?

According to the review, Google stopped displaying FAQ rich results in May 2026. Question-and-answer content itself still showed positive effects on sessions and AI citations, but the markup no longer earns the expanded SERP snippet. Product and Review schema still produce rich results.

How do you get cited in AI search results?

The evidence consistently favors content that answers the question directly, includes specific statistics, quotes and cites credible sources, shows clear authorship, and uses natural language rather than repeated keywords. Keyword stuffing reduced visibility in the GEO benchmark.

Is AI-generated content bad for SEO?

It depends on how it's used. Targeted AI-assisted enrichment of specific pages showed gains, while mass-publishing AI text across a whole site was associated with a 35% drop in organic sessions. Human editing of AI drafts outperformed raw AI text in controlled tests.

Can I trust SEO case studies?

Treat them as directional. The review found 64% of reported outcomes were positive and only 8% negative, a pattern that reflects what gets published rather than reality. Prefer controlled split tests, look for replication across independent sites, and run your own test before rolling a change out sitewide.

Sources

The meta-study: Krešimir Ćorluka, The SEO Meta Study: a systematic review of SEO, AEO and GEO case studies, Edition 1, September 2026. Ovisy, Zagreb.

Primary studies cited in this article:

• Pranjal Aggarwal et al., GEO: Generative Engine Optimization, arXiv 2311.09735
• Ahrefs, Republishing content
• Authority Hacker, Content update case study
• Backlinko, Content relaunch and Increase website traffic
• ContentPulse, Content freshness over six months
• Conversion Digital, Winning Google's AI Overviews
• Dave Ashworth, Keywords in URLs case study
• Deepak Gupta, Generative engine optimization for B2B SaaS
• Digital Surfer, CosmediTour case study
• Featured, Impactful SEO results
• Flento, Local SEO case study
• Fueler, Schema markup and organic traffic
• getAISO, Schema markup experiment
• HubSpot, AEO case study
• Inflow, Content refresh and Financial services SEO case study
• Lilach Bullock, AI for SEO without penalty
• Lumar, Peak Ace case study
• Moz, Surprising SEO A/B test results, On-page SEO changes case study and ROI of content re-optimization
• Neliti (academic), AI-based automated content generation
• Postion, Internal linking case study
• Rank and Discover, Structured data: 40% more traffic
• Reboot Online, AI vs humans content experiment, GEO examples and Outgoing link experiment
• Search Engine Land, Google to no longer support FAQ rich results
• SearchPilot, Top brands FAQs, Enriching page content, Most surprising tests of 2025, Authorship and E-A-T signals, Emoji title tags, Testing the canonical element, Canonicalising to product variations, A year of SEO testing and SEO moneyball
• Seer Interactive, Author bylines and AI Overview citations for an outdoor living leader
• seoClarity, FAQ schema CTR test
• SEO Mafia Club, Internal linking experiments
• SEO Melbourne, Linen House case study
• SEOTesting, BT customer story
• SEO Works, Reversing a Google penalty
• WordLift, Legal services SEO case study
• Writer Sanctuary, Schema markup case study
• Yext, Accor and Sofology
• Zyppy, Google disavow caution and Google disavow experiment