How I Research, Write & Use AI

Why This Journal Exists

The Journal is where I publish ideas, explanations, research-backed essays, working models and lessons from the things I am building or trying to understand. A piece should earn its place by giving a reader something more useful than a recap of information that already exists elsewhere: a clearer model, a meaningful distinction, original experience, a defensible synthesis, a practical implication, a decision aid, or a question worth carrying forward.

That standard matters more as publishing becomes easier to automate. The ability to produce text quickly is not a reason to publish it. I use software, automation and AI extensively, but the purpose of the Journal is still human: make something accurate enough to trust, distinct enough to be worth reading and useful enough that I would be comfortable putting my name on it.

Search visibility, AI citations, social distribution and newsletter reach are distribution outcomes. They may shape how a useful piece is packaged, titled, structured, linked or maintained. They are not sufficient reasons for the piece to exist.

Who Is Responsible for the Work

Journal articles are published under my name, Lorri Brewer. The byline identifies the person accountable for the published viewpoint, editorial standards and decision to make the work public. It does not imply that every sentence was manually typed without software assistance.

I use modern tools in the editorial process, including AI systems and automated agents. Depending on the article, those tools may assist with research planning, source discovery, synthesis, outlining, drafting, editing, data work, coding, metadata, internal-link analysis, accessibility work, media production or publication checks. Some articles may involve relatively little automation; others may use it heavily.

AI systems are not listed as authors. They do not have personal experience, responsibility or standing to make a first-person claim on my behalf. When an article contains a personal experience, observation, preference or story, that material must come from information I supplied, authorized or previously made public. An agent is not permitted to invent a memory, opinion, credential, relationship, quote or experience because it would make a story read better.

The practical rule is simple: tools can participate in production; accountability does not get delegated to the tool.

Research, Sources and Evidence

Research-backed articles are expected to have a traceable evidence base. The exact source mix depends on the question, but the default preference is to move toward primary and authoritative evidence whenever it is available: original research papers, official datasets, filings, legislation, standards, technical documentation, first-party reports, direct statements, public records and other sources close to the underlying fact.

Secondary reporting, analysis and commentary can be valuable for context, interpretation, disagreement and discovery. They should not quietly replace a primary source when the primary source is available and material to the claim.

AI-generated summaries are never treated as evidence. A model can help locate, compare or summarize potential evidence, but the underlying source has to carry the factual weight.

For newer Journal work, external evidence is normalized into canonical source records and connected to the article sections, research objects or datasets that depend on it. Reader-facing bibliographies and machine-readable citation relationships are derived from the same provenance model rather than maintained as separate lists that can drift apart.

Evidence should be proportionate to the claim

Not every sentence requires a citation. Personal interpretation, clearly framed opinion, ordinary connective reasoning and stable common knowledge do not need to be burdened with footnotes. Claims that are consequential, surprising, quantitative, current, disputed, technical or central to the argument deserve stronger evidence.

When a claim depends on a specific number, dataset or calculation, the goal is to preserve enough provenance that the path from source to conclusion can be reconstructed. Derived comparisons are treated as factual claims too. Units, time periods, denominators, assumptions and rounding should remain compatible with the source data.

Current facts need current verification

Facts that can change are checked against current sources as close to publication as practical. That includes things such as prices, product behavior, laws, policies, executive roles, market data, technical documentation and recent events.

An older source can still be useful for history or a durable concept. It should not be presented as if it proves a current condition that may have changed.

Uncertainty belongs in the article

A clean answer is not always an honest answer. When credible sources disagree, evidence is incomplete, a causal claim is uncertain or a forecast is inherently conditional, the article should preserve the boundary rather than smooth it away.

The job of editing is to make uncertainty understandable, not invisible.

How AI and Automation Are Used

AI is part of my editorial tooling. I use it because it can materially improve the breadth, speed and consistency of research and production when it is constrained by real sources, explicit standards and verification.

Depending on the assignment, AI may help with:

• turning a broad question into a research plan;
• finding potential primary sources, counter-evidence and missing perspectives;
• comparing large sets of source material;
• extracting candidate facts for verification;
• mapping relationships between ideas, entities and prior Journal work;
• testing an argument for gaps, unsupported leaps or obvious counterexamples;
• generating outlines or alternate structures;
• drafting or revising prose under an established voice and evidence set;
• checking repetition, grammar, readability and internal consistency;
• writing code for data processing, charts, diagrams or site functionality;
• preparing metadata, structured data inputs and internal links;
• running deterministic validation and publication workflows.

The amount of assistance varies. There is no useful honesty in pretending every AI-assisted article was produced by the same workflow or with the same percentage of human keystrokes.

The standard is outcome- and provenance-based: is the published work useful, original, accurate to the available evidence, clear about uncertainty and honest about how first-person material entered the piece?

AI output is not a source

A language model can produce a plausible sentence that is wrong, misdated, oversimplified or attached to a source that does not support it. For that reason, model output is treated as a working artifact, not as authority.

Load-bearing factual claims should ultimately rest on source evidence, a transparent calculation, an authorized first-hand account or clearly marked analysis. Citations are expected to support the claim they are attached to, not merely discuss the same topic.

What automation is not allowed to do

The editorial system is specifically designed to prevent a few failure modes that become easy at scale.

It must not fabricate sources, quotations, data, credentials, interviews or personal experiences. It must not turn private memory or unrelated connected information into publication authorization. It must not disguise sponsored or financially interested material as independent analysis. It must not create dozens of thin pages merely to cover query variations. It must not use a confident tone to erase a material evidence gap.

Automation can increase production capacity. It does not lower the publication standard.

Editing, Criticism and Publication

Substantial Journal work goes through more than a drafting pass. The current production system separates research, synthesis, writing, editing, critical review, packaging and release validation so that one confident generation is not treated as finished work.

For automation-ready articles, the working state is preserved in the repository so another qualified agent can inspect the question, evidence, unresolved gaps, editorial decisions and current review state without depending on a hidden chat transcript.

A separate critical-reading pass is used to challenge the complete composition. That critic may itself be an AI model or editorial agent. Its role is to find reader-level problems, unsupported logic, weak explanation, repetition, confusing structure or claims that need another research pass. A critic model is not an independent factual authority; its findings still have to be resolved against the evidence and the article itself.

The repository also runs deterministic checks that can surface or block specific classes of defects. Depending on the article, those checks cover content structure, provenance, unsupported structured claims, source links, duplicated prose, editorial media requirements, accessibility, schema output, search metadata, type/build integrity and other production contracts.

Automated writing-quality detectors are treated as signals rather than taste authorities. A grammar or readability tool can identify a real problem, produce a false positive or recommend a change that would make a sentence less accurate. The editor is expected to make that distinction.

Some Journal work is eligible for repository-policy publication after the required editorial and technical gates pass. That means an article can move through an automated release path without a final manual publish-button click. Automation of the final action does not change the standards that the release candidate has to satisfy.

Experience, Analysis and Opinion

First-hand experience is valuable precisely because it cannot be reconstructed from the public web. When I write from experience, the article should make that perspective legible without converting personal memory into universal evidence.

Personal observation can support statements about what I saw, did, built, tested or learned. Broader claims still need an appropriate basis.

Analysis and opinion are allowed to be analysis and opinion. The goal is not to wrap every judgment in false objectivity. The goal is to distinguish interpretation from externally verifiable fact strongly enough that a reader can tell what kind of claim they are evaluating.

Images, Charts, Data and Synthetic Media

Visuals have to do a job. Editorial photography can provide evidence, context, atmosphere or human texture. Charts and diagrams should clarify a relationship that is harder to understand in prose. Decorative graphics do not become evidence because they look analytical.

When a chart uses data, the dataset and source lineage should be preserved. The visual should use compatible units, sensible scales and enough labeling that the reader can understand what is being compared.

Stock photography is credited when the source information is available through the production workflow.

AI-generated or materially synthetic media may be used where it is appropriate, but it should not be presented in a way that reasonably invites a reader to mistake it for documentary evidence of a real event, person, place or measurement. When synthetic origin is material to how the image or media should be interpreted, that origin should be disclosed.

Corrections, Updates and Maintenance

Publication is not the end of the lifecycle. Articles are enrolled in a review cadence based on how quickly the subject can change, and evidence-based triggers can bring a piece forward for review sooner.

A review does not automatically mean rewriting the article. The right outcome may be no change, a metadata update, better internal links, a source refresh, a substantive correction, a major rewrite, consolidation with another page or retirement when the page no longer deserves to exist.

When a material factual error is confirmed, the article should be corrected rather than preserved for the sake of a clean publication history. The page’s modification date is updated when appropriate. If the nature of the correction is itself important context for the reader, the correction should be noted in the article rather than silently buried.

Minor copy edits that do not change meaning do not require a public correction notice.

Commercial Relationships and Conflicts

The site may discuss products, services, organizations, projects or other work in which I have a material interest. That experience can be useful source material, but readers should be able to tell when I have a material interest in the thing I am discussing.

When a Journal piece contains a sponsorship, paid placement, affiliate relationship, material ownership interest or other financial relationship that a reasonable reader would want to know about when evaluating the piece, it should be disclosed in context.

A commercial relationship does not automatically make a claim false or a product unworthy of discussion. Hiding the relationship would make the reader’s evaluation harder, which is why the disclosure matters.

Search, SEO and AI Discovery

I want useful work to be discoverable. The site therefore uses conventional search-engine optimization, structured data, semantic HTML, internal linking, descriptive metadata, crawlable public pages and a maintained sitemap.

Search can also inform editorial work by revealing questions people are asking, terminology they use, gaps in an existing corpus and changes in how a topic is being framed. That evidence can improve a commission. It does not get to manufacture the commission on its own.

I do not maintain a second lower-quality version of an article for search engines or AI systems. The visible page is the primary published work.

I also do not treat common “AEO” or “GEO” tactics as magic ranking switches. There is no special markup that guarantees an AI citation. The practical strategy is to make the underlying work worth retrieving: clear authorship, current facts, distinctive experience or analysis, direct answers where they help, strong source provenance, useful media, crawlable pages and a technically healthy site.

Scaled commodity content, fake third-party mentions, keyword variants with no distinct reader value and machine-only claims are outside the editorial standard.

Machine-Readable Provenance

The site exposes structured data to help search engines and other machines understand what readers can already see.

Published Journal articles use BlogPosting structured data with a canonical Person author, publication and modification dates, article sections, source citations and related entity or dataset relationships where the content supports them. The author and site point back to this page through Schema.org publishing principles.

The public research system also exposes crawlable provenance surfaces for sources, datasets and reusable research. Those support pages are generally marked noindex so they can provide traceability without competing with the main Journal article in search results.

Structured data is generated from canonical content and evidence records. It is not a hidden layer for claims that the visible page cannot support.

The same principle applies to AI discovery: machine readability should make real editorial truth easier to retrieve, not create a second truth for machines.

Where These Standards Apply

These standards govern substantive Journal publishing on LorriBrewer.com and the public provenance attached to it. Other surfaces can have different production requirements. A short social post, product interface, transactional email, legal document, private note or raw project update does not need to pretend it went through the full Journal pipeline.

The standard should be proportionate to the consequence and promise of the content.

The page itself will change as the publishing system changes. When tooling, review gates or disclosure expectations materially change, this policy should be updated with them.