Fix access before entity, entity before content, and content before reach. A fetcher that can't reach your page makes every later fix invisible, so the order you work in matters as much as the list of things to fix.
This is the practical follow-on to the full list of what the agent does on your site: once you know what it checks, the question that actually changes your week is which of those checks to work on first.
Because each layer depends on the one before it being read at all. An engine that never fetches the page can't see your Organization data, however clean it is. An engine that fetches the page but can't tell who you are has no reason to trust your answer-first copy over a competitor's. And a page that answers the question perfectly still loses to a competitor with more reviews and mentions backing it up. Fixing a later layer before an earlier one is broken is work that doesn't show up yet.
Proof: OpenAI documents its fetchers by name and user agent, GPTBot for training, OAI-SearchBot for its search index, ChatGPT-User for in-conversation browsing, each one either reads your page or does not: OpenAI — GPTBot, OAI-SearchBot and ChatGPT-User.
Why it matters: a site with perfect answer-first copy and a blocked robots.txt looks, to every engine, exactly like a site with no copy at all.
Whether an AI fetcher can actually retrieve your page: not challenged by a CDN, not blocked in robots.txt, not marked noindex. This is the layer that makes everything else possible or pointless. Three places it breaks most often: a robots.txt rule written for search crawlers that accidentally disallows an AI one, a CDN or bot filter (Cloudflare's "Block AI bots" setting is the common case) challenging a fetcher that robots.txt allows, and a noindex tag left over from staging.
| Access problem | Free check | Who fixes it |
|---|---|---|
| Robots.txt disallows an AI fetcher | Does your robots.txt let AI crawlers in? | The agent's switch, reversible |
| CDN or bot filter challenges the fetcher | Is your CDN or bot filter challenging AI crawlers? | A person, in the CDN's own settings |
| A page is marked noindex | Is a priority page marked noindex? | The agent proposes, you decide |
Do this first: run the free AI crawler access check before touching anything else on this list.
Whether an engine that reaches your page can tell, without guessing, who you are. This is Organization schema in the page head, a sameAs link to your other profiles, and a Wikidata item if you have one. Two businesses with a similar name and a plain logo read as ambiguous to a model, and an ambiguous entity doesn't get recommended, it gets skipped for a clearer one.
Proof: schema.org defines sameAs as the property for linking an entity to its other authoritative pages, the mechanism engines use to resolve who you are across sources: schema.org — sameAs.
| Entity gap | Free check | Who fixes it |
|---|---|---|
| No Organization data in the page head | Does your site say who you are in structured data? | The agent's switch, reversible |
| Schema doesn't link to your other profiles | Does your schema link to your profiles elsewhere? | The agent's switch, reversible |
| No Wikidata item | Do Wikipedia and Wikidata know your brand? | A person, it's a one-time outside submission |
Bottom line: entity work is mostly a few structured facts stated once and kept consistent, not an ongoing effort.
Whether the page itself actually answers the buyer's question in a form an engine can lift. That means the first sentence gives the answer rather than a warm-up, each section stands on its own without needing the rest of the page for context, and the page carries a visible heading structure, a date and an author. This is the layer readers notice too, which is why it tends to get fixed before access even though it does the least good until access is already clean.
| Content gap | Free check | Who fixes it |
|---|---|---|
| Page doesn't open with the answer | Do your pages open with the answer? | Words the agent writes, you approve |
| Sections depend on the rest of the page | Can each section of your page stand on its own? | Words the agent writes, you approve |
| No visible heading, date or author | Do your pages carry headings, dates and an author? | The agent's switch or a quick template fix |
Why it matters: the agent can draft this layer, but it never publishes a word without your approval, the same rule that applies to every page edit it proposes.
Whether anyone outside your own site backs up what you're saying: reviews, press, comparison articles, other pages linking to the ones that matter. An engine weighs a claim more when it sees the same fact repeated on a page you don't control. This is the one layer Reflexa doesn't pretend to automate: getting listed in a comparison article or earning a review takes a person, a pitch, or time, not a switch.
| Reach gap | Free check | Who fixes it |
|---|---|---|
| No independent press or reviews | Is there independent, citable material about your brand? | A person, judged by a person too |
| Comparison articles don't mention you | Do your external profiles agree on who you are? | A person, outreach and consistency |
| Priority pages aren't linked from the rest of the site | Are your priority pages linked from the rest of the site? | The agent's switch or a template fix |
Honest framing: reach is the layer with no shortcut. The other three can be fixed in a sitting; this one is earned over weeks.
Work the four layers in order, not by whichever feels most urgent:
Bottom line: the free check at dashboard.reflexatechnologies.com/audit runs all four layers in this order and tells you, by layer, which one is actually holding you back.
Access first: make sure ChatGPT, Claude and Perplexity's fetchers can actually reach your pages. Nothing else, entity data, answer-first copy, reviews, does any good if the fetcher never gets in.
Because a blocked fetcher never reads the page at all. Clear Organization data and an answer-first lede only help once an engine has actually retrieved the page, so fixing them before access is checked wastes the work.
No, and it says so plainly. It flips four reversible switches itself (fetcher access in robots.txt, Organization data, the sitemap, llms.txt), writes page text for your approval, and marks reach work, reviews and press, as something only a person can do.
No. The agent proposes removing a noindex tag and flags why, but it never flips that switch on its own. You decide, because noindex can be intentional.
Run the free checks in the same order: AI crawler access, what AI sees, schema and entity, then llms.txt. Each one tells you, by layer, what failed and the fix.
The free check runs access, entity, content and reach in this exact order and shows you the evidence for each one. 3 minutes, no card.