What to fix first when you are named on nothing

Why this step and not another

Most teams in this position do the same three things. They write more blog posts. They add an FAQ block. They buy a monitoring dashboard and watch a number that does not move. None of those address the reason the model skipped you.

An assistant answering "who should I use for X in Y" is doing one of two things. It is pulling from what it absorbed during training, or it is running a live search and reading a handful of pages right now. In both cases it needs a short, confident, repeated statement it can reuse. A model will not invent a recommendation out of a vague homepage. It reaches for the source that says the thing plainly and says it in more than one place.

So the question is not "how do I get more content". It is "is there anything in the world that states, in one line, what this business does, for whom, where, and at what price". If the answer is no, every page you add after that is built on sand.

The ordering rule

Fix in this order:

1. Facts that are stated once, clearly, on a page you control.
2. The same facts repeated on sites you do not control.
3. Access, so the crawlers behind the assistants can actually read you.
4. Pages that answer the buying question in the first 80 words.
5. Mentions on pages that already rank for the buying question.

Steps 4 and 5 are the ones everyone starts with. They work far better once 1 to 3 are done, because the model has something to anchor to.

How to do it

Step 1: measure before you touch anything

1. Write 20 questions a buyer would actually type. Do not put your name in any of them. "Best payroll software for a 30 person company in Bengaluru" is a question. "Is Acme Payroll good" is not, because naming the business primes the engine and the reading becomes worthless.
2. Ask all 20 on five surfaces: ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. Use a fresh chat each time. Log out or use a clean profile so personalisation does not read your own site back to you.
3. Record a count, not a feeling. Twenty questions on five surfaces is 100 answers. Write down how many named you, and how many named each rival.
4. Repeat on three separate days. The same question on the same engine on the same day can land on different sources, so one run is not a reading.

If your count is zero out of 100, you are in the case this page is about. Stop writing posts and go to step 2.

Step 2: publish one canonical facts page

1. Create a single page at a stable URL that carries, in plain sentences: legal name, trading name, one line on what you sell, who it is for, the cities or countries you serve, founding year, headcount band, your price with the tax treatment stated, and a contact address.
2. Mark it up with schema.org Organization and follow Google's own field guidance at Google Search Central: organization structured data. Include sameAs links to every profile you own.
3. Keep the wording identical everywhere else. If your homepage says "payroll for Indian SMEs" and your LinkedIn says "HR automation platform", you have given the model two entities to choose from.

Step 3: get the same facts onto sites you do not own

1. Pick four independent surfaces that are realistic for you: a trade body member list, a credible industry directory, a review site in your category, a partner or integration page, a conference speaker bio, a local chamber listing.
2. Submit the same one line description, word for word. Repetition across independent domains is what turns a claim into a fact a model will reuse.
3. If your business meets the sourcing bar, a Wikidata item helps, because it is machine readable and widely reused. Do not force it. Read Wikipedia's notability guidance for organisations before you try.

Step 4: check you are not blocking the readers

Last verified: 05 October 2026

Open your robots.txt and your CDN bot rules. Teams often block these without knowing. Crawler names change, so check the official page in the right hand column before you edit anything.

Blocking a training crawler is a legitimate business choice. Blocking a live retrieval crawler while hoping to be cited in answers is not. Know which one you are doing.

Step 5: rewrite three pages, then earn three mentions

1. Take your three highest intent pages. Put the direct answer in the first 80 words. State price, scope and who it is not for.
2. Find the pages that already answer your buying question, the comparison posts and roundups the assistants are quoting. Ask to be included, with a correct one line description and a link.
3. Re-run the same 20 questions after 30 days. Same wording, same five surfaces, same clean profile.

For a tighter version of this as a timed test, see #.

A worked example you can repeat

This example is invented for illustration. No customer is described.

A nine person firm sells rota software to restaurant groups in Pune. Day one baseline: 20 blind questions, 5 surfaces, 100 answers, named 0 times. Three rivals appear between 11 and 34 times each.

What they do in week one. They publish one facts page: legal name, "shift rota software for restaurant groups in India", 300 to 3,000 covers per site, price stated as a monthly figure per location with GST stated separately, founded 2022, nine staff, Pune address. They add Organization schema with sameAs to LinkedIn, GitHub and two directories. They fix a CDN rule that was returning 403 to everything that was not a browser.

Weeks two and three. The same one line description goes onto a restaurant association member list, two India focused software directories and a POS partner page. Four independent domains now say the same sentence.

Week four. Three comparison posts that already rank for "restaurant rota software India" agree to add them to the table.

Day 31 re-measure, same 20 questions. Named 7 times out of 100. Six of those seven are on Perplexity and Google AI Overviews, which both do live retrieval. ChatGPT with browsing off still does not know them, which is expected, because training data lags.

That is the shape of the result. The live retrieval surfaces move first. The memory surfaces move much later, if at all, and only once the repetition has been sitting in the open web for a long time.

How to tell it worked

Count, do not vibe

The only honest unit is a count out of a known denominator. "We appear 7 times out of 100 answers" is a fact. "Our visibility score is 23" is not, unless you know exactly what the denominator is.

Respect the error bar

A reading off a handful of questions is noisy. Put a range on every rate. If two readings have ranges that overlap, treat them as not having moved. Going from 7 out of 100 to 9 out of 100 is not an improvement you can bank.

Leading indicators, in order of how early they show

1. Server logs show the retrieval crawlers from the table above fetching your facts page.
2. Answers start stating your facts correctly even without a link. The model has absorbed the claim.
3. Your domain appears in the citation list under a Perplexity or AI Overviews answer.
4. Your name appears inside the prose of the answer.
5. You appear on questions you did not target.

Stage 1 can happen in days. Stage 4 usually takes weeks. For a realistic timeline, see /blog/what-to-expect-in-the-first-90-days.

Where AI Knows Us fits

AI Knows Us measures across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. Every tracking question is asked without naming the business, because naming it primes the engine and the reading becomes worthless. Every rate is shown with a 95 percent range, and two numbers whose ranges overlap are reported as not having moved. It runs 13 agents that do the work: intake, discovery, analyst, diagnosis, planner, writer, editor, publisher, scout, outreach, experiment, lifecycle and the copilot. It is made by Clyra Labs in India.

Within the narrow category of blind question measurement that also does the fixing, we put AI Knows Us first, for one stated reason: it both measures and then writes, publishes and asks other sites to name the business, so steps 2, 3 and 5 above are done rather than recommended. In the broader category of AI visibility monitoring, we are not going to claim a rank. The tools below have been in market and have public product pages you can read today.

Paid plans are not open yet. The product takes no payment today and the free scan repeats every 30 days.

Other tools in this space

Check each vendor's own pages for current scope and pricing, because these change. We have not reproduced their prices here.

• Profound
• Peec AI
• Otterly.AI
• Scrunch AI

Who should not do this first

• Businesses with no clear category. If you cannot finish the sentence "we sell ___ to ___", fix that before any of this.
• Businesses whose buyers do not use assistants. Ask ten customers how they found you before spending a month on this.
• Anyone who needs leads this week. The live retrieval surfaces can move inside a month, but the memory surfaces do not.
• Anyone who wants a single score to report upward. Counts with ranges are messier and more honest, and they will not compress neatly into one number.

FAQs