A new site gets impressions but no clicks because appearing in a result set and winning the click are two separate judgements the search engine makes, and the first happens far faster than the second. Impressions mean the index has your pages and considers them relevant. Clicks mean it considers you one of the best few answers, a ranking decision that authority and time feed into. A wall of impressions with almost no clicks is not a failure. It is what the first two weeks of a working site look like.
Ross Jones, Founder, The Hopium Lab. Last modified 22 July 2026.
A classic-search field report from running this across several zero-authority sites of my own, with the numbers. This is search-engine optimisation, not language-model citation, the two are different games and I will be clear about which one the data supports.
How fast can a brand-new site start showing up?
Days, not months, far faster than the timelines everyone quotes. A domain with no backlinks and no history starts appearing in search results, on page two or three for real queries, within a couple of weeks of publishing content the index can read.
The numbers from my own properties are the point of this post. A consumer price-comparison site I run went from zero to 10,800 impressions in roughly nine days from a standing start, sitting at an average position of about 22. A separate enterprise data platform I run, Integri.us, sat flat near zero for two months, then stepped into a climb the day a proper content structure went live, same domain, same everything except the pages, a clean before and after in one search-console graph. Neither ranks well yet. Both are unmistakably showing up, which a fortnight earlier neither did.
Showing up is the first half of the battle, and it arrives in days. A new site can be visible in a week. It simply is not visible in a good position yet, and those are not the same thing.
Why do impressions arrive before clicks?
Impressions arrive first because they are gated by relevance, and clicks are gated by rank, two different bars, cleared at two different speeds. The index will show a new, relevant page to searchers almost immediately, parked at whatever position its thin authority earns, which is usually page two or three. It will not send clicks until that position climbs into the range where people actually look.
So the honest early graph is high impressions and a click-through rate that looks alarming. One of my sites sits at a 0.5% click-through rate against those ten thousand impressions. Every guide frames sub-1% click-through as a problem to fix with better titles. At an average position of 22 it is not a title problem, it is a position problem, and click-through rises mechanically as position improves, with no change to the page at all. The impressions are the leading indicator. The click-through rate at position 22 is telling you your rank, not your writing.
| Signal | What it measures | Speed on a new site | The common misread |
|---|---|---|---|
| Impressions | Indexed and judged relevant | Days | Ignored because clicks are low |
| Average position | Where you sit in the results | Weeks to months | Expecting page one at once |
| Click-through rate | Whether you are a top answer | Follows position | Read as failure at position 22 |
| Clicks | The actual payoff | Slowest | Quitting before they arrive |
What is the method that produces the impressions?
The method is one page per question, each page structured so the index can read exactly what it answers. Three moves do most of the work, and none of them is a trick.
Make the page title the literal question a person types, not a marketing headline, a question-shaped title matches a question-shaped query. Answer that question in the first sentence, completely, so the answer stands on its own. And serve real HTML: the answer has to be in the page source, not painted in by JavaScript after load, or the crawler may never see it. The sites that took off did nothing clever. They stopped writing a homepage and started writing answers, one page per question, each one true on its own line.
That last point is where most new sites quietly fail before they start. A beautiful landing page ranks for nothing, because nobody searches your value proposition. An impressions graph is filled by distinct question-pages, each matching a distinct real query, a carousel is not an index entry.
The fastest way to fill a search-console impressions graph is to stop writing pages about yourself and start writing pages that answer the questions your buyers already type.
Is getting cited by an LLM the same thing?
Not the same, but the foundation is shared, and that is the part worth being precise about, because it is where the hype and the over-correction both live. The data above is classic Google search: impressions and position in the ten-blue-links index. It does not prove a language model will cite you, and I will not pretend it does. But the on-page work that produced it is the same groundwork an assistant needs, because a page a crawler cannot parse is a page an assistant cannot quote.
So the foundations pull double duty. Clean server-rendered HTML, a question-shaped title, a standalone first-sentence answer, a linked primary source, those make you legible to the search index today and legible to a model tomorrow. Nail them now and the search payoff is immediate and measurable; the language-model payoff is set up rather than delivered.
What the foundations do not do on their own is complete the LLM half, and the evidence on that is worth knowing. A study of Product Hunt startups found on-page optimisation had roughly no correlation with whether an assistant surfaced a brand, while referring domains and community presence did. A large brand-mentions analysis put mentions and links well ahead of on-page factors. Adding schema markup produced a small decline on one AI-answer surface in a controlled test, and the serving-file convention vendors sell for this gets fetched a negligible number of times against hundreds of millions of bot visits.
Nail the foundations now: they win classic search immediately and they are the precondition for language-model citation later. Just do not mistake the foundation for the finished building, the LLM half is completed off your site, by references you have to earn.
What are the failure modes?
Four, and each is a working method abandoned or misread.
Quitting at the impressions plateau. The graph shows ten thousand impressions and fifty clicks, someone reads it as failure, and the site is abandoned in week three, the week the index finished accepting it. The impressions were the win.
Optimising click-through at position 22. Rewriting titles to lift a click-through rate that is low only because the rank is low. The fix for position 22 is time and authority, not a better snippet. The effort is spent on the wrong number.
Buying markup as a shortcut. Adding schema and serving files because a vendor framed them as the route to AI citation. Add schema because it helps classic rich results, then stop expecting it to do a job the evidence says it does not do.
Writing for the homepage. Pouring effort into a landing page nobody searches for, while the question-pages that would actually fill the graph go unwritten.
How do you do it, and how do you read the result?
Ship one page per question, structured for the crawler, and read the graph for the right signal over ninety days.
GET INDEXED AND SURFACED FAST (CLASSIC SEARCH)
The Hopium Lab · v1.0 · 22 July 2026 · take it, fork it, argue with it
PER PAGE
[ ] Title is the literal question someone types, not a headline
[ ] First sentence answers it completely, true with no context around it
[ ] Server-rendered HTML, the answer is in the source, not painted by JS
[ ] One page, one question. Do not bundle five questions onto a page
THE SITE
[ ] Sitemap submitted to Search Console on day one
[ ] robots allows crawling, check it, do not assume it
[ ] Enough distinct question-pages to give the index a topic to map
READING THE GRAPH (where everyone goes wrong)
[ ] Week 1-2: impressions climbing, clicks near zero = WORKING
[ ] Low CTR at position 20+ is your rank, not your title. Leave it alone
[ ] Position is the number to watch move, over weeks
[ ] Do NOT expect clicks before position reaches the first page
WHAT THIS DOES NOT DO
[ ] It does not get you cited by an LLM. That is off-site and slower
[ ] Do not buy schema or serving files as an AI-citation lever
THE TEST: impressions rising, clicks not? You are winning the first half.
Call that failure and you will quit in the week it started working.
The discipline is patience with the correct number. Impressions rising is the search index telling you it has accepted the site, and it comes fast enough to be genuinely encouraging once you know to look at it rather than at the click count. Clicks are the payoff and they come last, after position, after time. Show up first. Everything commercial is downstream of showing up, and showing up is the part nobody tells you is quick.
Ross Jones, Founder, The Hopium Lab. Last modified 22 July 2026. The figures are from sites I run.