AI Disclosure

In plain English. TuBe SEO is AI software, and we would rather show you the machinery than describe it vaguely. Models research topics, draft articles, generate imagery, score drafts, analyze competitor pages, sample answer engines and write outreach pitches. Every one of those outputs passes a person before it reaches your website, your inbox or a journalist. Models do not decide anything: nothing publishes automatically, no email sends without approval, the software has no code path to change a redirect, canonical or indexing directive, and no price or contract term is ever set by a model. AI also gets things confidently wrong, so every statistic we generate is tied to a named source, every draft is graded against a 33-check rubric, and a human editor signs off. After that you still need to check the facts about your own business, because no model can verify those.

1. Why we publish this page

The TuBe SEO platform uses artificial intelligence as a core part of how it works. Large language models research topics, draft articles, score drafts, analyze competitor pages, summarize reports and write outreach pitches. Generative image models produce the hero imagery on those articles. This is the production line, not an add-on with a marketing label on it.

This page names the module, the job the model does inside it, and the human control on top. It is written for the client deciding whether to buy, the lawyer reviewing what we do with the client's content, and any regulator who asks.

It is incorporated by reference into our Terms of Service and referenced by our Privacy Policy. Where the two conflict on a legal question, the Terms of Service control.

2. Where AI is used across the platform

Generative AI appears in seven places in the TuBe SEO platform. In six, the model is a production tool whose output a person reviews. In the seventh (AI visibility), the AI systems are what we are measuring, and their answers are the data rather than a draft.

Where	What the model does	What a person does
Research and fact gathering	Reads live sources, extracts claims, attaches a named source to each statistic	Sets the brief; owns the source list
Article drafting	Writes a 3,000-5,000 word draft to an outline with per-section budgets	Edits, corrects, approves
Hero imagery	Generates editorial illustration, typesets the headline into it	Accepts, regenerates or replaces
Scoring and suggestions	Grades a draft against a 33-check rubric, proposes additive fixes	Accepts or rejects each suggestion
On-page analysis	Reads the page and the pages outranking it, produces gaps and two scores	Applies changes by hand in the CMS
AI-visibility sampling	Is the subject: 6 answer engines are queried, their answers recorded	Approves every prompt before sampling
Outreach and reports	Drafts pitch copy from a scraped article; writes prose around report numbers	Approves every email before it sends

Conventional software does the rest: the job queue, rank tracking, crawling, geogrid scans, index checks, cost accounting, and every arithmetic operation in a report.

3. Research briefs and fact gathering

Before anything is written, TuBe SEO runs live web research on the topic. A model reads current sources, pulls out the claims that matter, and attaches each statistic to a named source. That brief (the facts, the sources, the questions real searchers ask) is what the drafting step is allowed to work from.

The order matters. A model writing from memory produces plausible numbers it cannot support; a model writing from a sourced brief is doing a different and much safer job. Every statistic in a TuBe SEO article is attributed to a named source gathered by that research step, and an editor can follow the attribution back.

Models also cluster keywords and topics, grouping a seed term by which pages rank together and grading competitor gaps by winnability. Volume, difficulty, cost-per-click and rank positions come from search and SEO data providers, not from a model.

4. Long-form article drafting

TuBe SEO drafts long-form articles with a large language model, from an outline that sets a word budget for each section. Sections are written in batches rather than one pass, so quality holds to the last word instead of degrading down the page. A finished draft runs 3,000 to 5,000 words with question-phrased headings, self-contained sections, direct answers up front, an FAQ block and full structured data.

Drafts are produced against each client's own brand voice, keyword registry and guardrails, held separately from every other client's. The model writes to those constraints; it does not choose them.

A draft is a draft. It is scored, edited and approved before it becomes a deliverable, and approving an article and publishing it are two separate actions taken by a person.

5. Generative imagery

The hero images on TuBe SEO articles are generated by AI image models. They are produced as editorial photography with a locked film-grade look, the article's headline typeset into the image, and the client's logo composited on. Regenerating an image keeps the headline, so a library's visual identity stays consistent.

Be clear about what these are: illustrations, not documentary photographs. They do not depict a real person, place, product unit or event, and TuBe SEO does not present them as though they do. We do not generate images of identifiable real people without documented authorization, and we do not pass synthetic imagery off as a photograph of a client's premises, staff or work. Where a client needs a real photograph, the client supplies it.

Some image models embed provenance metadata, such as content credentials or an invisible watermark. We do not strip it.

6. SEO and answer-engine scoring, and the suggestions we generate

Every TuBe SEO draft is graded live in the editor against 33 checks. Twenty-two match the full applicable RankMath test set (Basic, Additional, Title Readability and Content Readability), using RankMath's exact criteria and its pass, warn and fail grading. Eleven more cover answer-engine structure and content quality RankMath does not measure. A thirty-fourth runs only when secondary keywords are set. The groups are basic, additional, title, content, aio and quality.

The answer-engine checks decide whether a page gets quoted rather than merely ranked: an answer-first opener of 40 to 75 words, a comparison table, a numbered or bulleted list, an FAQ section, specific data or original numbers, and at least two authoritative citations to .gov, .edu or .org sources. The quality checks cover human tone (limiting AI cliches and em dashes), no placeholder markers left in the text, an average sentence of 20 words or fewer, a reading grade between 6 and 10, and exactly one H1.

Some checks are arithmetic a model has no part in: counting H1 tags, measuring sentence length, detecting a table. Others require judgment, and there a model makes the call. The writer can disagree and ship anyway. When the platform offers a fix, the fix is additive only. It never rewrites the writer's prose.

7. On-page competitor analysis and the Information Gain score

For a page flagged as underperforming, TuBe SEO reads the page and the pages ranking above it, then reports the topic gaps, the structural differences and the questions real searchers are asking. A model does the reading and the comparison. On the module's first live run, 1,095 pages were bucketed in seconds: 84 flagged for work and 47 marked protected.

Two scores come out. SERP Parity measures how closely a page matches what is already winning, with a target band of 70 to 85 and an over-optimization warning above it, because chasing 100 is counterproductive and the tool says so. Information Gain measures the opposite: the unique value the page carries that competitors do not, which is what answer engines reward when choosing a source to cite. Both are model judgments about text. When there is not enough evidence to score a page honestly, the platform says so instead of inventing a number.

Suggestions are tagged by safety tier and reviewed one at a time, Accept or Reject. Pages already on page one are marked PROTECTED and receive only additive suggestions (alt text, an FAQ block, schema), because a re-evaluation after edits can cost a top ranking for months.

8. AI-visibility sampling, where the AI is the subject and not the tool

AI Visibility is the one module where the AI systems are what we are measuring rather than what we are using to produce something. TuBe SEO queries 6 answer engines (ChatGPT, Claude, Perplexity, Gemini, AI Overviews and AI Mode) with buyer questions built from the client's keyword warehouse, and records three facts per prompt: whether an AI answer appeared, whether the client was cited as a source, and whether the client was mentioned by name.

A human approves every prompt before sampling spends anything. Sampling costs money per query, and the prompt list is the measurement instrument: a sloppy prompt produces a number that looks like data and is not.

Answer engines are non-deterministic. Ask the same question twice and the answer can differ. So results carry statistical confidence ranges rather than one-off screenshots, and a single favorable answer is never presented as a result. The Mention Gap report lists every question where AI answers name a competitor and not the client. The AI geogrid pins an AI Overview check to each point of the map-pack coordinate grid: a default 5 by 5 grid of 25 nodes, with a 7 by 7 ceiling of 49. These are samples of third-party systems we do not control, and nothing here promises any engine will cite anyone.

9. Outreach drafts and report summaries

TuBe SEO drafts backlink and digital PR outreach with a model, and no draft leaves the building without a person approving it. Each pitch references the prospect's actual published article, scraped from the live page and never invented, runs under 150 words, and makes one soft ask. Every email sits in a review queue until a human approves it. The send engine warms mailboxes gradually, respects reply and unsubscribe signals absolutely, and pauses itself if deliverability health dips.

In reporting, models summarize what changed across rankings, AI citations, content output and link growth. The numbers in a report are not produced by a model. They come from the data pipeline: rank checks, index checks, citation samples, verified placements. The model writes the sentence around figures it did not compute. Every report exports to CSV so you can check the arithmetic. The job queue also logs the exact cost of every AI and data call per client, so AI spend is auditable rather than estimated.

10. Where AI is not used

The limits matter as much as the capabilities, and several are enforced in the code rather than by policy.

Nothing is published to a client site automatically. The platform publishes to WordPress as native Gutenberg blocks and can deliver by webhook or to GitHub, but approve and publish are separate steps and a person takes both.
No email is sent without human approval. Every outreach message waits in a review queue. There is no autopilot mode.
No redirect, canonical tag or indexing directive can be changed by the software at all. These are detect-only by design. The platform will tell you a canonical is wrong; it has no code path to change one. That is an architectural decision, not a setting.
No pricing or contractual decision is made by a model. Scope, price, terms, renewal and termination are decided by people and written into a Statement of Work.
No consequential decisions about individuals. We do not use AI to make or materially assist employment, credit, housing, insurance, education or healthcare decisions.
No emotion recognition and no biometric categorization, anywhere in the platform.

If you ask whether you are speaking to a person or to software, you will get a straight answer.

11. Human review: four approval gates and one kill switch

AI output is a draft, not a deliverable. The control structure is four approval gates and a kill switch, and it is the reason we are comfortable putting generative AI this deep into a production process.

Prompt gate. AI-visibility prompts are approved by a person before any sampling runs or spends anything.
Article gate. Articles are reviewed before publishing, and approving an article and publishing it are two separate actions.
Suggestion gate. On-page suggestions are reviewed one at a time, Accept or Reject, and applying an accepted change is a manual act in the CMS.
Outreach gate. Every outreach email is approved by a person before it is sent.

Above all four sits a master kill switch on the outreach send engine that only the owner flips. When it is off, nothing sends, regardless of what is queued or approved.

A named person at TuBe SEO is accountable for every published artifact (the article, the image, the email, the report), and where the Statement of Work provides for client approval, a named approver on the client side signs off as well. Review is substantive: reviewers check factual claims, brand accuracy, claim-substantiation and regulatory exposure, tone, originality, and conformity with the platforms' published guidelines. A cursory click-through is not review. This is also the mechanism we rely on where transparency law excepts AI-assisted content reviewed under a person's editorial responsibility. See Section 16.

12. Accuracy, confident errors, and what you need to check

Large language models produce confident errors. They invent facts, statistics, quotations, citations, product details, prices and regulatory statements, in the same fluent register they use when they are correct. Fluency is not evidence. Any honest account of this technology starts there, and a vendor who tells you otherwise is selling you something.

What TuBe SEO does about it:

Sourcing before writing. Every statistic in generated content is attributed to a named source gathered by the live research step, not recalled from a model's training.
A scored rubric. Every draft is graded against the 33 checks in Section 6, including a citation check requiring at least two authoritative citations to .gov, .edu or .org sources, and a check that no placeholder markers were left in the text.
A human editor before publication. The gates in Section 11 are not steps that get skipped when a deadline is tight.
Refusing to score without evidence. Where the platform cannot support a number honestly, it says so rather than producing one.

No review process catches everything, and we do not claim ours does.

That leaves a category only you can close. Clients should review every deliverable for factual accuracy about their own business, because no model can verify it. Service areas, licenses and credentials, warranty language, prices, product specifications, staff names and titles, hours, awards, years in operation, and any claim regulated by a professional body or the FTC have to be checked by someone who knows the answer. Deliverables are not legal, medical, financial or other professional advice. Section 7(c) of our Terms of Service sets out the approval window.

13. Which model providers process your content

To do the work, TuBe SEO sends prompts to third-party AI providers. We describe them by category rather than by brand, because the vendors change and a stale name in a legal page helps nobody. A current list of the providers that process Client Data is available on request at info@gotubeseo.com, and clients under a Data Processing Addendum receive advance notice of changes.

Category of provider	What we send them
Large language model providers	Research topics, page and competitor text, keywords, brand voice guidance, outlines and drafts, report data, scraped prospect article text
Generative image model providers	Image prompts, headline text, and the client logo to be composited
Answer engines sampled for AI visibility	The approved buyer questions: public queries, not client content
Search and SEO data providers	Keywords, domains, URLs and coordinates for volume, rank, backlink and map-pack data
Cloud hosting and database providers	Storage and processing of everything above

Prompts may incidentally contain personal information sitting inside the material being processed: a name in a byline, an email address on a contact page. We do not transmit credentials, payment card data or government identifiers to AI model providers, ever.

If any category of your content must not be sent to a third-party model at all, tell us and we will fence it off. Say so in writing and name the category: a product roadmap, unreleased pricing, customer lists, anything under a separate confidentiality regime. We will exclude that material from AI processing and handle it manually, and where that changes scope or cost we will tell you before, not after.

14. Training, retention, and what we do with your data

TuBe SEO does not use client content, client site data or Client Data to train, fine-tune or improve any general-purpose or third-party AI model. That is a statement about our own conduct and we stand behind it without qualification.

The next part is worded carefully, because it describes somebody else's behavior. TuBe SEO accesses third-party AI models through their commercial APIs, using enterprise or business tiers where they are available for the capability we need. As of the date at the top of this page, the terms governing that access provide that inputs and outputs submitted through those APIs are not used to train the provider's models. Providers typically retain API data briefly for abuse monitoring before deleting it, and some offer zero-retention configurations for eligible endpoints; we use those where available. We do not use consumer-tier AI products for client work.

Those are commitments made by the providers, not by TuBe SEO, and they are subject to terms the providers can change. We do not knowingly permit client content to be used to train third-party models, we monitor those terms, and we will update this page if a material change occurs. What we will not do is tell you a third party's behavior is something we can guarantee, because it is not.

Separately, TuBe SEO may use de-identified and aggregated technical, SERP and performance data to improve our own prompts, workflows and scoring, on the terms in Section 12.7 of our Terms of Service. We do not train models on identifiable client content.

15. Ownership and copyright of AI-assisted work

On full payment of the fees due under a Statement of Work, TuBe SEO assigns to the client all right, title and interest that TuBe SEO holds in the deliverables, excluding our own platform, software, models, prompt libraries and workflows. Section 12 of our Terms of Service sets out the assignment in full.

The current position of the United States Copyright Office, set out in Part 2 of its report on copyright and artificial intelligence dated January 29, 2025, is that copyright protects only material with sufficient human authorship; that material generated entirely by AI is not protected; that prompting alone, however detailed, does not by itself make an output a human-authored work; and that where a work combines human and machine contributions, the protectable part is the human contribution, including creative selection, coordination, arrangement and modification of AI output. Copyrightability is assessed case by case.

TuBe SEO deliverables are produced with substantial human direction. A person sets the topic and brief, directs the research, decides the structure, edits the prose, integrates and verifies facts, selects and arranges the material, and adapts it to the client's brand before anything is approved. That is the human authorship the Copyright Office describes, and our production records document it.

The consequence, stated straight: you should not assume every element of an AI-assisted work is separately protectable. We assign what we hold, and we do not warrant that any deliverable is protected by copyright, that you could enforce copyright in it against a third party, or that substantially similar output has not been generated for someone else. If you register a deliverable, you are responsible for disclosing AI-generated material where the Copyright Office requires, and we will provide reasonable information about our production process to support that filing.

16. Disclosure obligations and the transparency landscape

AI transparency law is arriving in pieces, at different speeds. An honest account of the parts that touch what we do:

The EU AI Act. Article 50 transparency duties apply from August 2, 2026. Article 50(4) requires deployers publishing AI-generated text to inform the public on matters of public interest to disclose it as AI-generated, with an exception where the content has undergone substantive human review and a person holds editorial responsibility. The approval gates in Section 11 are how we meet that exception, and where it does not apply, we label.
FTC guidance on AI claims. The FTC expects objective AI claims to be substantiated before they are made, and has been enforcing against unsupported AI claims in business-to-business marketing as well as consumer marketing. Our published numbers describe what the software does, not what it will do for you.
State AI transparency laws. A growing set of state laws requires disclosure that a person is interacting with generative AI, at minimum on clear request, and constrains AI used to manipulate or discriminate. Provenance laws in some states apply above user thresholds far larger than ours; we will meet them if we cross them.

Independent of what any law requires: TuBe SEO will disclose AI involvement in the content we produce whenever a client asks or a platform requires it. We do not present AI-assisted content as exclusively human-written, and we do not suppress AI-origin metadata a provider embeds in an output. White-label clients may present our work under their own brand, but may not instruct us to remove a disclosure the law requires.

17. What we will not do

Some of these are legal exposure and some are simply how we want to run a business. All are firm.

No fabricated statistics. A number in a TuBe SEO deliverable comes from a named source or it does not go in.
No invented sources. We do not cite a study, article, report or expert we have not actually located, and we do not manufacture links to give a claim the look of authority.
No fake reviews and no fake testimonials. Not for clients, not for ourselves. We do not generate review content, write testimonials attributed to people who did not give them, or publish a case study about work we did not do.
No impersonation of a real person in outreach. Outreach goes out in the name of a real person at TuBe SEO or, where an engagement provides for it, a real named person at the client who has agreed to it.
No AI-generated content published without human review. There is no configuration, plan tier or rush job in which this gate is turned off.
No AI-generated imagery presented as a photograph of a real client, place or event. Generated imagery is editorial illustration and is treated as such.
No content designed to deceive a search engine, and no synthetic media depicting a real, identifiable person without documented authorization.

If you see something from us that looks like a breach of this section, tell us and we will treat it as an incident rather than a complaint.

18. Governance, changes and how to contact us

This page changes when the platform changes or the law does, and the dates at the top are the honest ones. Material changes affecting client engagements are handled under Section 23 of our Terms of Service. We review provider terms, model selection and the approval gates on a recurring basis, and log AI incidents internally.

To raise a concern about AI use in anything we produced for you, email info@gotubeseo.com and describe what you are looking at. That includes a fact that looks wrong, a source that does not check out, an image you do not want used, a category of content you want excluded from AI processing, or a disclosure you want applied to published work. We aim to acknowledge within 3 business days and to answer substantively within 10 business days. If a factual error has been published, correcting it comes before explaining it.

SaaSassins Dev LLC, a Wyoming limited liability company, trading as TuBe SEO.
Email: info@gotubeseo.com
Web: gotubeseo.com/legal/ai-disclosure/

We do not publish a street address or a telephone number. Email to the address above is our designated channel for notices, questions and concerns about this AI Disclosure, and it reaches a real person.