GEO Lab #06: Content Optimization for AI Search — What Does the Evidence Say?
What works, what fails, and what remains unproven? An evidence review for indie developers and solo founders.
TL;DR — What should you actually do?
If you're an indie developer or solo founder, you probably don't have a dedicated SEO team. Every hour spent optimizing your website is an hour you could spend building your product.
So which content optimization tactics are actually worth your time?
We reviewed seven major GEO studies published between 2024 and 2026, covering controlled experiments, real AI citation data, end-to-end retrieval benchmarks, and a real-world website intervention.
Three findings stand out:
1. Relevance and useful information have the strongest support. Content that closely matches a user's question, and provides concrete facts needed to answer it, performs better in controlled citation experiments and is associated with more citations in real AI search data.
2. Popular formatting tricks have weak evidence. FAQ blocks, TL;DR boxes, structured data, and similar tactics have not demonstrated consistent independent improvements in real AI citations.
3. Some GEO rewrites can backfire. Controlled experiments show that generic content rewrites can reduce retrieval performance, even when they were designed to improve the final AI answer.
The biggest limitation: Very few studies establish that changing a live webpage causes a sustained increase in organic AI search visibility.
The evidence at a glance
Scroll across the table to see all columns.| Content optimization tactic | Evidence status | Worth trying? |
|---|---|---|
| Match content to real user questions | Supported by controlled and observational evidence | High priority |
| Provide complete, concrete product information | Supported in controlled product-selection tasks | High priority |
| Add relevant statistics and credible evidence | Positive results in some controlled experiments | Yes, when evidence is genuine |
| Keep factual information up to date | Recency cues work in some controlled tests | Yes, when content actually changes |
| Keyword stuffing | Negative results in tested environments | Avoid |
| Generic GEO rewriting at scale | Often ineffective; sometimes harmful | Avoid blind application |
| Formatting-only rewrites | No consistent citation advantage | Optimize for readability |
| Fabricated statistics | Demonstrated factual reliability risk | Avoid |
| Answer-first / TL;DR boxes | Independent AI visibility effect unproven | Optional |
| FAQ sections | No stable within-site citation association | Add for real user needs |
| Author pages and first-sentence attribution | Direct effect on AI naming unproven | Useful for identity clarity |
| Tables, short chunks, separate answer pages | No proven universal advantage | Use when structurally appropriate |
| Schema.org markup | Independent AI citation lift unproven | Follow standard SEO needs |
“Supported” refers to the tested conditions, not guaranteed results on ChatGPT Search, Gemini, or other live platforms.
1. How to interpret GEO evidence
Before examining individual tactics, we need to distinguish what researchers actually measure.
A website can succeed or fail at several different stages:
| Stage | What it means | Example metric |
|---|---|---|
| Discovery | A search system finds the webpage | URL discovery status |
| Retrieval | The webpage enters the search results or available context | Retrieval hit rate |
| Citation | The AI includes the webpage as a cited source | Citation rate |
| Mention | The AI names the product or creator | Mention rate |
| Recommendation | The AI actively recommends the product | Recommendation rate |
| Referral | A user clicks through to the website | AI referral sessions |
These outcomes are related, but they are not interchangeable.
For example, a study showing that a webpage contributes 40% more text to an AI answer does not establish that the webpage receives 40% more organic search traffic.
Four evidence levels
This review uses four categories:
Controlled experimental support: Researchers changed a specific factor while holding other conditions relatively constant. This provides evidence of an effect in the tested environment.
Real-world observational support: A tactic or characteristic is associated with more citations in actual AI responses. Other differences between websites may still explain the association.
Negative or inconclusive experimental results: A tested tactic failed to improve the measured outcome, or produced worse results, under specific conditions.
Insufficient direct evidence: The tactic is widely discussed, but its claimed effect has not been established through a sufficiently relevant experiment.
For every tactic, the key questions are: What was changed? What was measured? Which search stages were included? How well does the result transfer to a small product website?
2. Tactics with positive experimental evidence
2.1 Match your content to the questions people actually ask
Verdict: The most consistently supported content-level direction.
What does the evidence show?
A 2026 study, What Gets Cited, performed 252,000 controlled trials across six language models and 18 content factors.
Each trial compared two candidate sources differing in one factor. Topical relevance was one of the strongest determinants of which source received the first citation. [S3]
Another study, What Drives AI Citations?, analyzed 158,099 real AI answers containing 468,037 citations.
When comparing pages within the same website, researchers found:
| Page characteristic | Association with citations |
|---|---|
| Best paragraph matches the query, +1 standard deviation | +15.5% |
| Title matches the query, +1 standard deviation | +14.1% |
These are observational estimates, with differences across platforms and page types. They do not predict the percentage improvement from rewriting any particular page. [S5]
When does this work?
The controlled evidence applies when a model chooses between sources it has already received.
The observational evidence applies to real citations from ChatGPT, Gemini, and Perplexity in the study's commercial-query datasets.
Neither study proves that changing a page title or paragraph will independently cause the same increase in live search citations.
Evidence confidence: Stronger than most content GEO tactics for relevance as a factor; limited causal evidence for live website edits.
What should an indie developer do?
Start with the actual problem your user is trying to solve.
For example, suppose you're building a time-tracking tool for freelancers.
A generic page:
Productivity Tool — Features and Benefits
may communicate less useful information about the user's specific need than:
How to Track Time Across Multiple Long-Term Freelance Projects
The second page could explain the workflow, provide examples, show relevant features, and describe limitations.
The important improvement is genuine task relevance and useful content. Replacing titles with question-shaped keywords alone has no guaranteed effect.
2.2 Provide complete, concrete product information
Verdict: Supported in controlled product recommendation and comparison tasks.
What does the evidence show?
The same 252,000-trial study found that explicit pricing information consistently influenced first-citation selection.
Information completeness and some trust-related cues also produced smaller positive effects. [S3]
This is relevant because many AI search questions concern product decisions.
A user might ask:
- Which free tools support project-based time tracking?
- How much does a particular SaaS cost?
- Does this app work offline?
- What are the differences between two competing products?
A webpage that contains the necessary facts gives the model more information to use.
When does this work?
The evidence comes from controlled comparisons of already-provided candidate sources.
Pricing effects are especially relevant to commercial and product-selection questions. The result cannot automatically be generalized to every article or research topic.
Evidence confidence: Relatively strong within the tested product-selection setting; limited evidence for organic search lift.
What should an indie developer do?
Make the following information easy to find when relevant:
- Who the product is for.
- What problem it solves.
- Core features and practical workflows.
- Pricing, free limits, and subscription conditions.
- Supported platforms and integrations.
- Known limitations.
- Meaningful differences from alternatives.
Avoid exaggerated or unverifiable comparisons.
These improvements can help human users make decisions regardless of their measurable GEO impact.
2.3 Add verifiable data, references, and relevant quotations
Verdict: Positive controlled evidence exists, but results vary substantially across research environments.
What does the evidence show?
The original GEO: Generative Engine Optimization paper (KDD 2024) evaluated several methods for rewriting source documents.
Its controlled benchmark showed the following relative changes in Position-Adjusted Word Count:
| Tactic | Relative change |
|---|---|
| Add relevant quotations | +41.5% |
| Add statistics | +31.6% |
| Improve language fluency | +27.5% |
| Cite credible sources | +26.9% |
These are changes in the study's source-exposure metric. They are not increases in organic citation rate, product recommendations, or traffic.
The experiment used preselected sources and examined how their content contributed to generated answers.
Later research, including C-SEO Bench and SAGEO Arena, found much weaker or negative results for several rewriting strategies under different evaluation conditions. [S2][S4]
When does this work?
Evidence addition can influence answers when the modified source is already available to the model.
The benefit is less predictable when the modified document must compete in an actual retrieval and ranking process.
Results may also vary by topic. Statistics may be useful for research or performance comparisons, while direct quotations may be more relevant to interviews or historical accounts.
Evidence confidence: Moderate for specific fixed-source experiments; mixed for broader search visibility.
What should an indie developer do?
Publish information competitors cannot easily reproduce:
- Original benchmarks with clear methodology.
- Real performance tests.
- Documented user research.
- Product comparisons based on actual use.
- Case studies with explicit assumptions and limitations.
- Accurate statistics linked to their sources.
A reproducible experiment with a modest result is more useful than a dramatic claim with no supporting evidence.
2.4 Maintain accurate and genuinely current information
Verdict: Recency cues influence some controlled citation decisions.
The What Gets Cited study found that recent timestamps could increase the likelihood of receiving the first citation in its two-source comparisons. [S3]
However, the study measured model responses to date-related cues.
It did not establish that regularly changing the displayed publication date improves organic AI search visibility.
Evidence confidence: Moderate for specific recency-cue experiments; weak for artificial content-refresh strategies.
Practical implication: Update pricing, feature availability, supported integrations, compatibility information, and other time-sensitive facts when they genuinely change.
Use accurate publication and update dates. Artificially refreshing timestamps without meaningful changes has no established GEO benefit.
3. Tactics with negative or inconclusive experimental results
This section requires careful interpretation.
A negative result in a specific benchmark does not prove that a technique will fail on every platform.
However, these findings challenge claims that certain GEO practices are universally effective.
3.1 Keyword stuffing
Verdict: Negative results in early GEO experiments.
The KDD 2024 GEO paper tested keyword stuffing as a source modification strategy.
On its main benchmark, keyword stuffing performed worse than the unmodified baseline in the principal word-exposure metric.
In its additional Perplexity fixed-file evaluation, the same technique produced approximately 10% lower Position-Adjusted Word Count than the baseline. [S1]
When does this conclusion apply?
These results concern content inserted into controlled generative-answer settings.
They do not imply that using accurate product terminology or relevant search phrases is harmful.
Evidence confidence: Moderate for the tested negative result.
Recommendation: Use the terminology your users naturally use. Avoid mechanical repetition of phrases such as “best AI productivity tool” throughout unrelated content.
3.2 Applying generic GEO rewriting templates across a website
Verdict: Often ineffective in benchmarks, sometimes harmful to retrieval and citation.
Evidence from C-SEO Bench
C-SEO Bench (NeurIPS 2025) compared optimization approaches across question-answering and product recommendation tasks, multiple domains, and different competitive adoption scenarios.
Many tested content rewriting methods failed to produce consistent improvements and sometimes reduced document rankings.
The research also found that gains decreased when more competing documents adopted similar optimization methods. [S2]
Evidence from SAGEO Arena
SAGEO Arena (KDD 2026) addressed a limitation of fixed-source experiments.
It built a test environment with:
- 171,003 documents.
- 2,700 queries.
- Retrieval, reranking, and answer generation stages.
The optimized documents were reintroduced into the search corpus before evaluation.
In its principal experimental configuration, conventional body-text optimization methods had the following average relative changes:
| Search stage | Average change |
|---|---|
| First-stage retrieval hit rate | −9% |
| Reranking hit rate | −16% |
| Final citation rate | −6% |
The reported values are specific to the benchmark and its model configurations.
Some structural optimization methods performed differently, and the paper does not establish that all content rewriting is harmful.
Why might rewriting backfire?
A rewrite can increase the apparent usefulness of a document after retrieval while changing the features used to retrieve it.
A page may become more attractive to the answer generator but less competitive in the preceding search stage.
When does this conclusion apply?
The negative results come from controlled tasks and a constructed end-to-end search system.
They should not be interpreted as measured average losses across live ChatGPT Search websites.
Evidence confidence: Moderate to strong within the tested systems; generalization to commercial search remains uncertain.
Recommendation: Avoid site-wide automated rewriting solely to satisfy a generic GEO checklist. Prefer targeted changes that solve actual content deficiencies.
3.3 Formatting-only optimization
Verdict: No consistent independent citation advantage has been established.
Common suggestions include:
- Turning paragraphs into bullet lists.
- Replacing prose with tables.
- Using shorter paragraphs.
- Adding more headings.
- Reformatting content without adding new information.
The What Gets Cited study found relatively small or inconsistent effects from formatting-only changes in its controlled two-source experiments. [S3]
The real-world citation study likewise did not identify common formatting markers as consistently strong independent predictors once website-level differences were controlled. [S5]
What does this mean?
The evidence does not establish that these formats are universally ineffective.
Formatting can improve human readability, accessibility, and usability. It may also matter for particular model architectures or tasks.
The unsupported claim is that formatting alone reliably increases organic AI citations.
Evidence confidence: Moderate for the absence of a consistent effect in the studied settings; limited for universal conclusions.
Recommendation: Structure content according to what helps users understand it. Do not treat formatting changes as a guaranteed GEO intervention.
3.4 Adding invented statistics to appear authoritative
Verdict: Demonstrated factual reliability risk.
LIVE-GEO (October 2026) tested GEO rewrites published as real webpages.
Its Statistics Addition strategy produced 80 rewritten product descriptions that were all flagged by the study's factual-consistency checker. [S7]
This indicates a substantial factual reliability problem in that particular generated set. A checker flag does not automatically establish that every number was independently proven false.
The study did not establish a precise causal effect of inaccurate statistics on organic search visibility.
Evidence confidence: Strong evidence of problems detected in the tested outputs; unknown general impact on AI citation rates.
Recommendation: Never generate supporting numbers solely to improve GEO. Use actual measurements and disclose methods.
4. Popular tactics that remain insufficiently validated
These tactics are common in GEO advice, but the evidence does not yet support strong claims about their independent effect on organic AI visibility.
Some are still valuable for readers, search engines, or brand clarity.
4.1 Answer-first writing and TL;DR summary boxes
Popular claim: Putting the answer at the top makes AI more likely to cite the page.
What evidence exists?
A real-world Glasp study included changes to titles and opening summaries, but these were implemented alongside other interventions, including URL canonicalization and new content based on observed demand. [S6]
The What Drives AI Citations? study found that summary boxes were associated with more citations across different websites. However, the association was not statistically stable when comparing pages within the same website. [S5]
What remains unproven?
Whether independently introducing an Answer-first section on an existing webpage increases organic ChatGPT citation rate.
Worth trying? Yes, for long research articles, technical documentation, and complex tutorials when it improves reader comprehension.
GEO confidence: Low for an independent organic citation gain.
4.2 FAQ sections
Popular claim: AI search favors question-and-answer formats.
What evidence exists?
In the 158,099-answer observational study:
- Across websites, FAQ markers were associated with 18.1% more citations.
- Within the same website, the estimate was −3.4% and did not demonstrate a stable significant gain. [S5]
The difference matters: websites containing FAQs may already be stronger publishers for other reasons.
What remains unproven?
That adding an FAQ block to an otherwise unchanged product page causes an increase in natural AI citations.
Worth trying? Yes, when customers genuinely ask recurring questions.
GEO confidence: Low for an independent citation benefit.
4.3 Author profiles, bylines, and identity signals
Popular claim: Clear author information helps AI connect the work to its creator and mention the creator by name.
Possible implementations include:
- A dedicated author page.
- A visible byline.
- Consistent author identity across articles.
- Links to verified social profiles.
- Naming the creator in the opening sentence.
What evidence exists?
The large observational citation study did not find a stable independent citation benefit from author information in within-website comparisons. [S5]
But citation of a webpage and naming its author are different outcomes.
Existing evidence does not directly resolve whether these practices improve creator-name mentions after the creator's work is recommended.
Worth trying? Yes, if building a personal brand or research portfolio.
GEO confidence: Insufficient direct evidence for improved author naming.
This remains a particularly relevant research gap for independent creators.
4.4 Tables, chunking, and short standalone pages
Popular claim: AI prefers short, self-contained content chunks.
What evidence exists?
Formatting studies have not established a universally superior structure.
Google's official guidance also advises website owners not to pursue unnecessary content chunking specifically for its AI search features. [S8]
There is no generally validated optimal paragraph length or page length for AI citation.
Worth trying? Use tables for genuine comparisons, sections for distinct topics, and standalone pages when they serve separate user needs.
GEO confidence: Low for a universal content-chunking advantage.
4.5 Schema.org and JSON-LD
Popular claim: Structured data makes AI more likely to understand and recommend a webpage.
What evidence exists?
The 2026 observational citation study found no stable independent increase in within-site citations associated with Schema.org markup. [S5]
Google's official AI Search guidance does not require special generative-AI structured data. Existing supported structured data remains relevant to traditional Search features. [S8]
Worth trying? Implement accurate, relevant structured data following the target platform's official requirements.
GEO confidence: Low for a separate AI citation lift.
Scope note: Schema.org belongs primarily to technical and machine-readable page information, rather than prose content optimization.
4.6 Artificial freshness and aggressive content republishing
Popular claim: Updating dates and repeatedly republishing pages makes AI regard them as more relevant.
Controlled experiments show that date cues can influence some citation decisions. [S3]
However, no reliable general causal benefit has been established for repeatedly updating timestamps without substantive content changes.
Worth trying? Maintain genuinely current product information. Avoid artificial update histories.
5. What real-world GEO experiments tell us
Controlled studies help identify mechanisms. Real websites introduce additional challenges: crawling, indexing, competition, ranking, and platform changes.
Two studies are especially useful for interpreting these limitations.
5.1 Glasp: Large traffic growth does not equal a large optimization effect
Glasp studied actual ChatGPT referral traffic to its website.
Some pages received a group of AEO-related changes, while other pages on the same domain served as a contemporaneous comparison group.
From January to May 2026:
| Group | ChatGPT referral growth |
|---|---|
| Treated pages | 6.1× |
| Untreated pages | 3.5× |
The researchers' interrupted time-series analysis estimated an intervention-aligned level change of approximately 1.82×.
However, a conservative placebo-in-time test returned p = 0.16.
The authors therefore characterized the result as suggestive rather than conclusive. [S6]
What can we learn?
First, AI platform growth can create substantial traffic increases even without webpage optimization.
Second, a single before-and-after comparison can substantially overstate an intervention's effect.
Third, because Glasp applied several changes together, the study cannot tell us which specific content tactic caused any incremental improvement.
This study measures referral sessions, not simply whether AI answers cite a page.
5.2 LIVE-GEO: Real webpage effects remain sensitive to the evaluation setup
LIVE-GEO published product descriptions as real accessible webpages and tested several GEO rewriting strategies across AI products, including ChatGPT, Grok, and Gemini.
The study covered 80 fictional brands across 18 product families. [S7]
Some rewriting strategies performed well in product ranking tasks. But rankings varied by platform, presentation order, and candidate setup.
The critical limitation is that the experiments generally supplied candidate URLs or a catalogue page to the AI systems.
The study therefore tests real webpage reading and selection under constrained candidate discovery.
It does not fully represent an ordinary user asking an open-ended question without providing URLs.
What can we learn?
A tactic that works when a model is given candidate webpages may have little measurable impact on whether those webpages appear in natural search results.
Evaluation must account for the full route from discovery to retrieval to final recommendation.
6. A practical content optimization plan for small product websites
Based on the evidence reviewed, here is a reasonable allocation of limited time.
This is an evidence-informed recommendation, not a universally tested optimal workflow.
| Priority | Action | Why |
|---|---|---|
| P0 | Ensure each major page answers a real user question | Relevance has the most consistent supporting evidence |
| P0 | Publish accurate product facts, pricing, and limitations | Supported in controlled product decision tasks |
| P1 | Add original examples, benchmarks, and credible evidence | Positive but conditional experimental evidence |
| P1 | Maintain current and accurate information | Useful for users; some controlled recency evidence |
| P2 | Improve readability, headings, tables, and summaries | Strong usability rationale; independent GEO lift unproven |
| P2 | Add FAQs and author information when genuinely useful | Real user and identity value; GEO increment uncertain |
| P3 | Pursue specialized formats solely for AI visibility | Limited demonstrated incremental value |
Diagnose the problem before changing the content
If AI cannot find your page: Check technical access, actual indexing or discovery evidence, internal linking, and query relevance. Content formatting is unlikely to resolve every discovery problem.
If your page appears in search sources but is rarely cited: Examine whether it actually provides the requested information, distinctive facts, and supporting evidence.
If your product is cited but not named: Investigate whether the page clearly identifies the product and connects its features, claims, and identity. The effectiveness of particular identity tactics remains unproven.
If your product is mentioned but not recommended: Examine product suitability, competitive information, and user intent. Citation and recommendation require separate measurement.
If citations increase but conversions do not: Analyze referral quality, landing-page experience, and conversion behavior independently.
These diagnoses are practical hypotheses; they should be tested against actual observations.
7. How reliable is the research?
The seven core studies cover different stages of AI search. Their findings should be read together.
| Study | Research method | What it establishes best | Main limitation |
|---|---|---|---|
| [S1] GEO (2024) | Fixed-source controlled rewriting | Some content edits change generated source exposure | Does not test natural discovery |
| [S2] C-SEO Bench (2025) | Multi-task, competitive benchmark | Generic rewrites often fail or lose effectiveness | Tested tasks and simulated conditions |
| [S3] What Gets Cited (2026) | 252,000 paired source trials | Factors affecting first-citation selection | Two sources are already provided |
| [S4] SAGEO Arena (2026) | Constructed retrieval → reranking → generation pipeline | Some rewrites damage earlier search stages | Not a commercial live search engine |
| [S5] What Drives AI Citations? (2026) | 158,099 real AI answers, observational analysis | Real citation patterns and within-site relationships | Cannot directly establish edit-level causality |
| [S6] Glasp Natural Experiment (2026) | Website logs and contemporaneous controls | Real referral changes amid platform growth | Bundled treatments and time confounding |
| [S7] LIVE-GEO (2026) | Real webpages and deployed AI products | Sensitivity of content ranking under live reading | Candidate URLs supplied by researchers |
Three important research gaps
1. Organic search discovery
Many controlled GEO experiments begin after a source has already been selected or provided.
We still need better evidence that changing webpage content independently increases its chance of being found in natural AI searches.
2. Long-term, cross-platform effects
Models, retrieval providers, ranking algorithms, and citation policies change.
A result observed in one model version or benchmark may not hold across ChatGPT, Gemini, Grok, and Perplexity over time.
3. Small-site generalization
A strategy that works on a well-established publisher may have very different results for a new SaaS website with little external recognition.
The costs of obtaining reliable causal evidence are particularly high for independent developers with only a handful of pages.
8. Final recommendations
For indie developers and solo founders, the most defensible content GEO strategy today is to invest in information that is useful regardless of the search interface:
Understand the questions people ask. Answer them accurately. Provide concrete facts and original evidence. Keep the information current.
Treat formatting and identity enhancements as useful product-content decisions whose additional AI visibility benefits still require measurement.
Avoid optimizing blindly for AI-specific templates, fabricated authority signals, or dramatic visibility claims unsupported by real search data.
The research also suggests an important distinction:
What makes an already-retrieved document attractive to an AI model may differ from what makes the document discoverable in the first place.
That distinction explains why a tactic can succeed in a controlled GEO benchmark yet fail to improve visibility on a real website.
The bottom line
There is experimental evidence that content changes affect AI citation behavior.
There is much less convincing evidence that any single on-page content tactic reliably produces lasting increases in organic AI search visibility across platforms.
For small teams, that uncertainty should influence both what they optimize and how much time they invest.
The goal is to prioritize improvements with genuine informational value, measure observable outcomes, and remain skeptical of guaranteed GEO growth claims.
References
Primary research
[S1] Aggarwal et al. (2024). GEO: Generative Engine Optimization. ACM SIGKDD 2024.
[S2] Puerto et al. (2025). C-SEO Bench: Does Conversational SEO Work?. NeurIPS 2025, Datasets & Benchmarks.
[S3] Vishwakarma, Kumar & Jamidar (2026). What Gets Cited: Competitive GEO in AI Answer Engines.
[S4] Kim et al. (2026). SAGEO Arena: A Realistic Environment for Evaluating Search-Augmented Generative Engine Optimization. ACM SIGKDD 2026.
[S5] Chen & Koudas (2026). What Drives AI Citations? Content, Retrieval and Model Knowledge in Generative Engine Optimization (GEO). Preprint.
[S6] Watanabe & Nakayashiki (2026). Disentangling Answer Engine Optimization from Platform Growth: A Log-Based Natural Experiment on ChatGPT Referral Traffic. Preprint.
[S7] Kong et al. (2026). Do GEO Gains Survive the Live Web? Evaluating In-the-Wild Visibility Gains of Generative Engine Optimization via LIVE-GEO. Preprint.
Platform guidance
[S8] Google Search Central — Optimizing Your Website for Generative AI Features.
Review methodology and limitations
This is a focused narrative review, not a systematic meta-analysis.
The studies were selected for their relevance to on-page content changes and AI visibility measurements. We distinguish peer-reviewed research, preprints, controlled experiments, observational evidence, and official platform documentation.
The reviewed results do not establish universal effects for all models or industries. Newly published preprints require further independent replication.
Related but distinct tactics—backlinks, third-party comparisons, brand authority, broad distribution, and technical indexing infrastructure—fall primarily outside the scope of this Content GEO review. They should be evaluated separately.
Research scope: On-page content relevance, factual completeness, evidence, rewriting, formatting, author information, and closely related content presentation tactics.
Intended readers: Indie developers, solo founders, and small product teams evaluating how to allocate limited effort to GEO.