A passing re-scan proves the website check changed. It does not prove more citations. How to use both numbers without fooling yourself.
You fixed robots.txt, allowlisted ChatGPT-User, aligned the Organization name with the About page, and shipped author bios. Someone asks, "Did it work?" If you only watch brand mentions in ChatGPT for a week, you will fool yourself in both directions — false highs and false lows. Re-scan the site for the part you can prove.
A passing re-scan proves the website check changed. It does not prove more citations. Use both numbers without mixing them up.
What a BrandKnown re-scan can prove
BrandKnown’s readiness score is a website checklist: crawler access, brand/entity clarity, and related on-site issues — with copy-ready fixes. A scan takes about 60 seconds. The free first scan establishes a baseline; later scans show whether those checks flipped. For what the score is (and is not), see AI search readiness score explained.
| After a fix… | Re-scan can show | Re-scan cannot show |
|---|---|---|
| Allowed an AI bot in robots | Access check improved | That Perplexity will cite you |
| Matched H1 and schema name | Entity clarity improved | Share of AI answers |
| Unblocked WAF 403s | Fetch succeeds | Ranking in AI Overviews |
| Added Organization sameAs | Profile links detected | Revenue from AI referrals |
| Fixed soft 404 / canonical clutter | Cleaner discovery signals | Guaranteed branded lift |
Agency plan is $49/mo with 10 saved brands and 100 scans/month — enough to re-scan after batches of fixes without burning the month on vanity refreshes. That loop matches an agency AEO workflow for ten clients.
Why re-scan at all
- Confirm the fix landed in production. Staging wins do not count; DNS, CDN, and caching still serve old robots files.
- Catch regressions. A new Bot Fight rule can undo last week’s allowlist.
- Separate engineering done from marketing outcomes. Close the ticket when the check passes; open a different ticket for measurement.
- Keep client reporting honest. Agencies need a before/after that is not "the intern saw us mentioned once."
Pair this with sober measurement playbooks like tracking AI referrals in GA4 and measuring branded search lift.
What citations and referrals can (sometimes) show
- AI referral traffic in analytics (small samples lie).
- Branded query lift in Search Console (laggy, multi-cause).
- Manual prompt panels (bias-prone; treat as QA).
Similarweb / TechCrunch (June 2025) reported about 1.13B AI-platform referrals to top 1,000 sites in the measured month, up 357% YoY, with ChatGPT responsible for more than 80% of those AI referrals — while Google Search still sent about 191B. Growth is real; your site is not entitled to a proportional slice. Similarweb Gen AI Landscape 2025 (US desktop, Sep 2025) saw longer sessions (~15 min vs ~8) and modestly higher conversion rates (~7% vs ~5%) from ChatGPT referrals than Google in their cut — directional, not your forecast.
Pew Research (March 2025): AI Overviews on ~18% of searches; click on a traditional result 8% with an AI summary vs 15% without; click on a citation in the summary ~1% of visits. Even perfect readiness cannot invent clicks Google does not give.
Gartner’s February 2024 forecast that traditional search volume may drop 25% by 2026 is a forecast. Do not put it on a client slide as a measured KPI.
Step-by-step: fix → re-scan loop
- Run a baseline scan; export or note failing checks.
- Triage P0 access issues before copy tweaks.
- Ship a batch of fixes (robots, WAF, schema name, About).
- Purge caches; wait for CDN truth.
- Re-scan the same URLs/brand.
- Record: checks that flipped, checks that did not, residual risks.
- Only then sample assistants or analytics for outcome signals.
- Schedule the next re-scan after the next change batch — not daily for superstition.
Hypothetical: say you run 20 prompts before and after. Ten mentions become twelve. That is not significance. The re-scan that shows GPTBot went from blocked to allowed is the proof your engineering worked.
Cadence that respects the 100-scan budget
| Cadence | Use for | Avoid |
|---|---|---|
| After each P0 fix batch | Access / WAF / robots | Scanning every typo |
| Weekly during a 30-day push | Active client projects | Vanity score chasing |
| Monthly steady state | Regression watch | Ignoring regressions for a quarter |
| Per locale / brand | Multilingual or multi-brand | Assuming one homepage represents all |
Reporting language that stays honest
| Say this | Not this |
|---|---|
| "Access checks passed on re-scan" | "We are now AI Overview optimized" |
| "Entity name consistent across title, H1, schema" | "Guaranteed citations" |
| "AI referrals up; sample still small" | "AEO doubled revenue" |
| "Readiness score is a site checklist" | "Score predicts ChatGPT rank" |
For client-facing structure, reuse the spirit of the client AEO report template: score, access, identity, evidence, next steps — no ranking promises.
Mistakes
- Re-scanning before cache expires, then declaring the tool broken.
- Changing ten things at once with no notes, so a pass cannot teach you what mattered.
- Treating score theater as strategy — chasing 100 instead of fixing the blocked pricing page.
- Skipping re-scans entirely and judging success only by chat anecdotes.
- Equating a higher score with a Pew-style Overview citation (those ~1% citation clicks are a different system).
Bottom line
Re-scan to prove the website changed. Measure citations and referrals separately, with humility. BrandKnown exists to make the first part fast and concrete — ~60-second scans, free first scan, Agency at $49/mo. The ceiling remains: readiness is not a citation forecast. Use the score as a checklist, and sleep better knowing the bots you welcomed can actually see the brand you fixed.