Backbrief Site Check

Site Check sample: backbrief.ai

Real run of Backbrief Site Check 0.2.0 on 2026-10-13, on backbrief.ai, the owner's own site. The recheck follows once fixes are made.

The panel for this run is saved. It is not published on this page.

Run by Backbrief Site Check 0.2.0 (the shipped product, run from its own command and skill files) on the owner's own site. Date read: 2026-10-13. Everything below was read; nothing was edited, submitted, sent, bought or published.

Intake (answered from the record on the owner's GO of 2026-10-13, not as live answers from him)

1. URL: https://backbrief.ai
2. What the owner wants to know: why AI assistants do not cite backbrief.ai when someone asks for a way to grade a business idea before building it, check a website before launch, check an ad or email before it goes out, or check a contract before signing, using Claude Code; and what to fix.
3. Existing site. No Search Console export supplied, so content decay reads NOT CHECKED for that reason.
4. AI readiness intake question: yes. The site's support chat widget (HighLevel) is a scripted bot that generates replies shown to visitors (owner's statement).
5. AI-crawler intent: not stated. The Reach check on crawler intent reads NOT CHECKED; status, raw HTML and noindex ran.
6. Speed report: none supplied. PageSpeed Insights was queried once per page and returned HTTP 429 (quota), so speed reads NOT CHECKED.
7. GEO or AEO vendor claims: none.
8. Competitors: none named. The competitor set was fixed as the distinct third-party domains cited in the first repeat of each prompt (57 domains, listed in the panel file) and stays fixed for every later repeat.

Panel: market US, language English, repeat minimum 3 (product default), 5 prompts. Panel file: outputs/site-check/panels/backbrief.ai-panel-v1.md, answers in backbrief.ai-panel-v1-answers.md.

Read this before using section 4. The run could reach one of the two named engines, and not by the route the brief asked for.

• Perplexity: the 15 runs (5 prompts, 3 repeats) ran in a browser profile on the owner's machine, signed in to the owner's Perplexity account. They are not logged-out runs. The run then tried a logged-out context, and Perplexity refused it: it answered "Sign up and repeat your request." and redirected to a sign-in page. The run signed in to nothing. A signed-in route may carry personalization from the account's memory. P4 R1 shows it: the answer contained a local folder path the run never supplied. The sample is labeled signed-in for that reason. The 15 threads now sit in the owner's account history.
• ChatGPT: NOT CHECKED, 0 repeats. The Chrome profile on the owner's machine is signed in to the owner's ChatGPT account. Submitting a prompt from the logged-out page moved the session into that signed-in app. The run submitted one prompt (P1) into the owner's signed-in ChatGPT, then stopped. No answer was read. Whether a saved thread now exists in that account is NOT CHECKED, because there is no capture.
• The verifier gate (command step 3) was not run by this session, by instruction. The controlling session runs it. Stop-slop was run over the prose.

Summary

NOT CHECKED items carried into the overall verdict: ChatGPT engine (section 4); AI-crawler intent (2, 4); PageSpeed (2, 6); Search Console rows (2); schema validator (3); borrowed assets (1); accuracy of what engines say about the brand (4); vendor promise grade (4); content decay (5); 27 cited sources that could not be read (4); checkout amount (7). AI readiness sits outside the scheme and reads Not verified.

Why AI assistants are not citing you

Panel version 1, one engine (Perplexity web), 3 repeats per prompt, 5 prompts, run 2026-10-13. Directional.

Result: backbrief.ai was mentioned in 0 of 15 answers and cited in 0 of 15. All 5 prompts are gap prompts.

Gap prompts with no cause evidenced: none.

Top fixes, ordered by the prompts each targets:

1. Corroboration, 5 prompts (P1 to P5): the sources the answers cited are a research list; whether to be on any of them is an owner decision (fix 2).
2. Answer, 1 prompt (P2): no page of the site states, in a heading and the sentence after it, that Site Check checks a website before launch in Claude Code. The Site Check page never uses the word "launch" (fix 3).

Directional diagnosis from a panel of 5 prompts and 3 repeats. It shows conditions found on the site and in the cited sources. No fix is promised to change what any assistant cites.

Section 1. Design against the slop signature: fix

Pages read: / at 1280 px and 390 px (screenshots home-1280px.png, home-390px.png), /site-check at 390 px (site-check-390px.png), all in outputs/site-check/2026-10-13-backbrief.ai-assets/.

Section 2. Technical search eligibility: fix

Date read for every row: 2026-10-13. Raw captures are in ...-assets/raw/.

Section 3. Structured data: fix

Date read: 2026-10-13. Pages: the 9 sitemap pages, raw <script type="application/ld+json"> blocks.

Structured data is not presented here as a lever for AI citation. Google states: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add." (https://developers.google.com/search/docs/fundamentals/ai-optimization-guide, read 2026-10-13.)

Section 4. AI answers: baseline and citation gap: fix

Frame: panel v1, Perplexity web only, US, English, 3 repeats per prompt, run 2026-10-13 between 13:11 and 13:31 US Eastern. Directional. The route caveat at the top applies to every number here.

Prompts (verbatim, from query-and-prompt-research; sources in the panel file; no buyer-typed text was supplied, so the wording is the run's phrasing of the owner's own seeds):

• P1: How do I grade a business idea before I spend money building it? I use Claude Code.
• P2: Is there a Claude Code tool that checks my website before launch?
• P3: How do I check an ad, email or post before it goes out for unbacked claims and missing disclosures?
• P4: How can I use Claude Code to review a contract or proposal before I sign it?
• P5: I'm an independent consultant who prepares client website reports in Claude Code. What can I use to find out why AI assistants do not cite a client's site?

Results. Mentioned: 0 of 15. Cited: 0 of 15. Per prompt: P1 0 of 3 cited, P2 0 of 3, P3 0 of 3, P4 0 of 3, P5 0 of 3. A repeat of 3 is the minimum, and with every result at zero the one-run-variation caution matters less here than it would for a split result. The sources cited were stable in P1 and P2 (the same 10 URLs in all 3 repeats) and shifted in P3 to P5 (16, 11 and 11 of 25 URLs appeared in all 3).

Cited instead, per gap prompt (domains cited in all 3 repeats; full lists in the panel file):

• P1: aicodex.to, github.com (RiseMore-AI/startup-idea-validator), aitoolsclub.com, x.com, sellingwithnas.com, inc.com, skilldeck.dev, roborhythms.com, enigmatica.ai.
• P2: github.com (bzsasson/pre-launch-audit-skill), claudemarketplaces.com, support.claude.com, platform.claude.com, anthropic.com, academy.claude.com, producttalk.org, reddit.com, thebootstrappedfounder.com.
• P3: ftc.gov, consumer.ftc.gov, ecfr.gov, law.cornell.edu, kelleydrye.com, blog.promise.legal, businesslawsw.com, contemporarydigitalmarketing.com, modash.io, influencerdb.net, checkmyads.org, medicare.gov, reddit.com.
• P4: americanbar.org, code.claude.com, learn.microsoft.com, support.microsoft.com, support.xbox.com.
• P5: foglift.io, otterly.ai, code.claude.com, blog.sshh.io, futureinsidernews.substack.com, oltre.ai, linkedin.com, reddit.com.

Cause 2, Answer (check 11). The rubric's standard: a heading naming the topic, a direct statement of the answer within the next two sentences, in raw HTML, readable alone.

Cause 3, Corroboration (check 12). Of 145 unique cited URLs, 118 returned 200 and none contained the string "backbrief" (case-insensitive, raw HTML). 27 could not be read (17 returned 403, 10 returned 404) and are NOT CHECKED by URL: they include the americanbar.org pages, several ftc.gov and consumer.ftc.gov pages, fcc.gov, inc.com, help.openai.com, uk.practicallaw.thomsonreuters.com and pub.towardsai.net; the list is in ...-assets/raw/cited-results.txt (rows whose first field is not 200). Per prompt, readable cited pages without the brand: P1 9 of 9, P2 10 of 10, P3 24 of 24, P4 33 of 33, P5 42 of 42. Reproduction for one page: curl -sSL https://github.com/bzsasson/pre-launch-audit-skill | grep -ic backbrief returns 0.

Two caveats on this cause. First, the check is mechanical: a cited page without the brand is a Fix. For P3 and P4 most cited pages are regulator, law-firm and vendor documentation where a product mention is not expected, so the diagnosis is weakest there. It is strongest for P1, P2 and P5, where the cited pages are listicles, a skill repository, a marketplace and vendor comparisons. Second, the diagnosis is a condition found, not proof of why any assistant chose its sources.

Section 5. Content decay: NOT CHECKED

Existing site, and the owner supplied no Search Console export. Step that would run it: export Search Console performance data for two equivalent periods (this quarter and the same quarter last year).

Section 6. Conversion structure: fix

Page read: https://backbrief.ai/site-check (the page that sells Site Check), 2026-10-13, at 1280 px and 390 px.

Section 7. Page claims: fix

Scope: the home page and /site-check, plus the claims on /backbrief that carry numbers and the statements in llms.txt that restate prices and versions. /deal-check, /claim-check, /backbrief/how-to, refunds, terms and privacy were read for structure and were not ledgered. Read dates are 2026-10-13. "Own material" means the owner's own pages or product files, so a match shows consistency, not independent proof.

Verdict: fix. No claim a source shows false was found; rows 4, 6, 7, 12, 13, 14 and 16 are unverified claims a reader could act on. A claim about what any named place's law requires: none found on these pages.

Section 8. AI readiness

The owner states that the site's support chat widget (HighLevel) is a scripted bot that generates replies shown to visitors (owner's statement, from the record on the owner's GO of 2026-10-13).

Verdict: Not verified.

Say "expand the AI readiness section" for the full list of questions to take to a professional, written out one per line with the facts each one needs.

Fix list

Addressed to whoever built the site. Ordered by what each fix prevents. "Owner decision" items change what is public or remove a file; a rollback copy comes first. No fix is promised to change what any assistant cites, how a page ranks, traffic or sales.

1. Retired files still served (section 2, check 10). Owner decision. Remove /agent-os-kit.zip, /getting-started-with-claude.pdf and /start-simple-with-ai-checklist.pdf, or redirect them to /backbrief, after saving copies of all three. Remove the three matching Disallow lines once the files are gone. Re-check: curl -sSI https://backbrief.ai/agent-os-kit.zip | head -1 expects a 404 or 308 status line (repeat for the two PDFs).
2. Corroboration, AI cause. Targets P1, P2, P3, P4, P5. Owner decision. The sources the answers cited are the research list; this run contacted none of them, and the list is never an outreach list. Whether to be on any of them is earned or not pursued, with no paid placement and no manufactured mentions. The lists per prompt are in the panel file; the recurring ones are: for P1 aicodex.to, github.com/RiseMore-AI/startup-idea-validator, inc.com, sellingwithnas.com; for P2 github.com/bzsasson/pre-launch-audit-skill, claudemarketplaces.com, producttalk.org, thebootstrappedfounder.com; for P3 kelleydrye.com, modash.io, contemporarydigitalmarketing.com, checkmyads.org; for P4 code.claude.com documentation and americanbar.org (regulator and documentation pages, so weakest); for P5 foglift.io, otterly.ai, oltre.ai, blog.sshh.io, futureinsidernews.substack.com. Prompts, verbatim: P1 "How do I grade a business idea before I spend money building it? I use Claude Code." P2 "Is there a Claude Code tool that checks my website before launch?" P3 "How do I check an ad, email or post before it goes out for unbacked claims and missing disclosures?" P4 "How can I use Claude Code to review a contract or proposal before I sign it?" P5 "I'm an independent consultant who prepares client website reports in Claude Code. What can I use to find out why AI assistants do not cite a client's site?" Re-check, per source the owner chooses: curl -sSL <source URL> | grep -ic backbrief returns 1 or more once the brand is present (it returns 0 on all 118 readable cited pages today). This proves presence, not citation.
3. Answer, AI cause. Targets P2 "Is there a Claude Code tool that checks my website before launch?" Owner decision (it states what the product is for). Add to /site-check, in raw HTML, a heading that names the job and a first sentence that answers it in plain words, for example that Site Check checks a website before launch inside Claude Code. The page now frames the product around AI citation and never uses "launch". Re-check: curl -s https://backbrief.ai/site-check | grep -ic "before launch" returns 1 or more, and the heading sits within two sentences of the answer. This proves the passage exists, not that any assistant will cite it.
4. Duplicate title and h1 on / and /backbrief (section 2, check 8). Give each its own title and h1. Re-check: for p in "" backbrief; do curl -s https://backbrief.ai/$p | grep -o "<title>[^<]*</title>"; done prints two different titles.
5. Dead sameAs target (section 3, check 5). Correct or remove https://x.com/CharlesAshe20 in the Organization markup. Re-check: curl -s https://backbrief.ai/ | grep -o 'x.com/[A-Za-z0-9_]*' shows a handle whose profile loads.
6. Confirm the four prices in Product markup (section 3, check 4). Owner decision. Re-check: curl -s https://backbrief.ai/site-check | grep -o '"price":"[0-9]*"' prints the confirmed price.
7. Header and footer on the Site Check page lead to the $29 product (section 6, check 7). Give the page its own navigation, or relabel the links. Re-check: curl -s https://backbrief.ai/site-check | grep -c 'href="/backbrief#pricing"' returns 0.
8. No proof on the Site Check page (section 6, check 3). The planned replacement sample can fill this. Re-check: curl -s https://backbrief.ai/site-check/sample | grep -c "Why AI assistants are not citing you" returns 1 or more. If the new sample should be findable, also drop its noindex: curl -s https://backbrief.ai/site-check/sample | grep -c "noindex" returns 0.
9. Copy tells (section 1, check 6). Rewrite the hero, the footer repeat of the hero, and the "things that come after the plan" heading as set out there. Re-check: curl -s https://backbrief.ai/ | grep -o "Grade the idea. Get the plan." | wc -l returns 4 today. It returns 2 when the footer drops the repeat and the hero keeps the phrase (one visible occurrence, counted twice in the raw HTML), and 0 when both drop it.
10. llms.txt sentence "Nothing in the line is free" (section 2, check 12). It becomes true once item 1 is done. Re-check: item 1's re-check.
11. Unverified claims (section 7, rows 4, 6, 7, 12, 13, 14, 16). Supply a source for each or remove it. Re-check: each row's wording present or absent in curl -s <URL>.
12. Cover-the-logo (section 1, check 5). A design opinion: the hero headline could be any idea-validator's. A headline that names Claude Code or the council would pass the test. No deterministic re-check; re-run the test by eye.

Not fixes: owner decisions and notes

Items 13 and 14 are a request to the owner and a note on how to re-run, not changes to the site. They keep the numbers 13 and 14 so the list reads 1 to 14 in sequence, and the 12 fixes above them are numbered 1 to 12. Item 12 has no deterministic re-check, so it sits last among the fixes.

13. Inputs that unlock NOT CHECKED rows. State the AI-crawler intent (sections 2 and 4); run PageSpeed on / and /site-check after the quota resets (sections 2 and 6); give Search Console access or an export (sections 2 and 5); run the live URLs through the Rich Results Test (section 3). These are requests to the owner, not site changes.
14. Re-run on a logged-out route. Section 4 rests on a signed-in Perplexity profile and no ChatGPT runs. A recheck that adds ChatGPT or uses a logged-out route is a new panel version and a new baseline, so /aeo-recheck against v1 compares like for like only on Perplexity in the same profile.

Questions for a licensed professional

Each is written so a professional who never saw this report can answer it. Places are named only as facts about the audience, and every point is an unverified local-law assumption; nothing here states what any place requires.

1. A site operated from Charleston, South Carolina shows visitors a chat widget that the operator describes as a scripted bot generating replies, labeled "Automated assistant; a person answers every support case." The operator has not said where visitors are located. Does that wording suffice as a notice for visitors in each place the site is offered, and what should it add? (From section 8.)
2. If the operator's practice does not match "a person answers every support case" in some instances, what exposure does the sentence create, and how should it be worded? (From section 8.)
3. The widget passes typed text and an email address to a CRM supplier, and the privacy page says page-view data is reported to an advertising platform. Is the privacy page's description sufficient for visitors in the places the site is offered? Place not stated. (From section 8.)
4. The site sells downloads under the line "All sales are final" with "Statutory rights unaffected" in the refund policy. Do those words give buyers in the places the site is offered the notice they need, and are they enough on their own? Place not stated. (From the home page and /refunds, read for wording only and not ledgered in section 7.)

Disclosure

This report was produced with the help of an AI model running inside a Backbrief checker. It is not legal, tax or financial advice, and no licensed professional has reviewed it unless the report names that review.

A pass means no confident weakness was found against the checks that ran. It is not a guarantee of ranking, citation, traffic or sales, and a check that did not run says so.

Run record

2026-10-13 | site-check | fired: yes | opened: no | owner answered yes (from the record on his GO of 2026-10-13); verdict Not verified; site backbrief.ai; overall verdict fix; verifier gate not run by this session

Appended to outputs/site-check/ai-readiness-runs.md.

Get Backbrief, the base

This is a sample report of a real run on 2026-10-13. Backbrief Site Check is available for $79, paid once.