SEOLift: your Search Console data, turned into this week’s work

A reading tells you what is broken. Search Console tells you what people search for and where you appear. Neither, on its own, tells you which page to spend Thursday afternoon on. SEOLift joins the two, and every kind of data you already have turns into something you can do:

• You have positions and impressions for every search, so you can see the pages stuck on page two for searches people actually make, and the one change holding each back: a title, a few internal links, a heading.
• You have your own click-through rate at every position, so you can find the pages that rank and get ignored, and win the clicks without ranking any higher.
• You have every search mapped to the pages it lands on, so you can catch two of your pages splitting one search, or the wrong page ranking for it, before you write a third page on the same subject.
• You have the questions people type, so you can put the question in a heading and the answer right under it: the change that earns the click and the quote in an AI answer.
• You have searches you show up for with no page behind them, so you can write the few pages your demand already asks for, in the words people typed.
• You have sixteen months of history, so you can win back lost traffic with the cause named from your own rows: overtaken, the result page changed shape, or people stopped searching.
• You have a before and an after, next to pages nobody touched, so you know for certain whether a change worked.
• You have the share of your demand the data can actually see, so you know which numbers are partial before you act on them.

The Goals screen sorts every opportunity into the outcome it serves, with the estimated gain for each and how many of them need no new backlinks.

• Page two to page one: pages ranking eleventh to twentieth for searches people actually make.
• Top ten to top three: pages already on page one, where a better title or a few internal links move the needle most.
• More clicks, same ranking: pages that rank well and are ignored in the result page, usually a title or description problem.
• Win back lost traffic: pages whose clicks fell against their own history, with the likely cause named.
• One page per search: two of your pages splitting the same search, or Google showing a page that is not about it while the page that is sits unlinked.
• Answer the questions people ask: how, why and which-is-better searches your pages rank for without asking them in a heading.
• Pages worth writing: searches you already appear for with no page on the site about them. These carry no click forecast, because a page that does not exist yet has no history to count from.

The Goals screen opens with a short note on your data: how much of your pages’ impressions the search rows explain (Google withholds rare searches, so a part always stays hidden), how much is your own brand, and how much lands on addresses the site does not offer as pages, such as redirects or noindexed URLs. When the rows see less than 60% of the demand, the note says the lists are partial.

Under it sits This week: five changes at most, best first, each with a rough effort from fifteen minutes to a day, and a line saying how many more are waiting and why they can wait. At the bottom is what we looked at and are not recommending: pages outranked on referring domains and searches nobody makes any more, listed so you know we saw them.

Open an opportunity and you get the same seven parts every time: what we found, which data says so, why it matters, the exact change to make, the expected result as a range, how sure we are, and what would prove us wrong. The action is one of a short vocabulary: rewrite the title and H1, write a meta description, add internal links, ask the question in a heading and answer it first, add structured data, fix the technical cause, update the content, redirect or merge, write a new page, or earn referring domains first.

Where the change is text, a suggested title, H1 or description is drafted for you. A model writes it; code checks it. Length, width in the result page, coverage of the search, keyword repetition, the brand suffix, and one rule above all: every number in the draft must appear in the page’s own text. A draft that fails twice is shown as a failure, with the reason, and the rules stay as strict as on the first try. A new page comes with a brief instead: the searches it has to answer, in the words people typed.

Nothing is changed on your site. You approve an opportunity, dismiss it with a one-line reason, or tell it you have made the change. From there an experiment starts: the page is compared with similar pages of the same template that nobody touched, so the season and the news cancel out and what you see is the effect of your edit.

The Experiments screen shows each change, its effect, a 90% interval, what it was compared with, and a verdict: Confirmed, Probable, No effect, Cost traffic, Inconclusive or Too early. The unflattering verdicts are the point. A tool that only reports its wins teaches you nothing about when to doubt it.

A measurement needs a reading to have seen the change first. If your edit is newer than the last reading, the page says so and offers to order one. A merge is looked for on the page that was merged away: a redirect, a canonical pointing at the page keeping the search, a noindex, or a rewrite that stops it competing. A new page is not measured as an experiment, since it has no before; the imports that follow show whether the searches move to it.

The module comes with the Business and Agency plans. It is off for every site until you turn it on, because reading a Search Console property is not something to start behind your back.

1. Connect a Search Console property to the site, on the site overview or in site settings.
2. Open SEOLift settings and switch on "SEOLift on this site". Search Console is read nightly from then on.
3. Add your brand terms, so searches for your own name are not mistaken for opportunities.
4. Set the market: a country and a language. Without a country the paid analysis stays off for the site, because a result page is only meaningful for one place.
5. Wait for history. No rule fires with less than ninety days of Search Console data; the first ninety days are read at once, and thirteen more months are filled in behind the scenes.

Search Console data is free to read. The second pass, which looks at the actual result page, checks keyword volume and counts referring domains, costs money per look, so each plan carries a monthly allowance of external checks per site. One check is one result page, one keyword batch or one backlink lookup. A cached answer costs nothing and counts for nothing: what a result page looked like yesterday is a public fact and is paid for once.

Settings shows what has been spent this month. Extra packs of a thousand checks are workspace-wide, and a site past roughly a hundred thousand pages draws three times the allowance, because it is three times the job.

In SEOLift settings there are two lists. What the site is about: when you fill it in, these words replace the subject read from Search Console, and the competitor gap keeps only searches that share one of them. Never show: a search holding one of these words, or every word of a phrase, disappears from every SEOLift list, from the week to pages to write.

Words are matched by their base form, so a term covers its inflections. The lists update within a minute of saving, without spending checks, and anything you allow again comes back on its own.

An empty list is a real answer, and the module says so in as many words: no page on this site is close enough to winning to be worth your afternoon. A page whose competitors outrank it on links is told so, with the numbers, and given no forecast. A search nobody makes any more is left alone. Both are recommendations we are glad not to make.

A search that is not your business, such as a stray phrase a page picked up by accident, is dismissed with a one-line reason and does not come back. If you leave the plan, nothing is deleted. Reading stops, and a measurement in progress pauses and may finish without a verdict.

A search you already appear for with no page about it used to be one line on a list. It is now a brief: the search to answer, the searches that go with it, a proposed address, the page the new one must not duplicate, and one sentence on why the current content does not cover it. The outline is in the words people search with, the questions they already ask become headings, and every fact is either already on the page or marked as needing a source.

A comparison search, such as X versus Y or an alternative to X, gets a comparison template with blank cells for price, features, complaints and when not to choose. They stay blank until a person fills them, and each shows as unchecked until then. Nothing is published and nothing is written for you, and a page that does not exist gets no click forecast, because there is no history to count from. Once you publish and give the brief the address, the next reading checks that most of the approved headings landed and names any number in the new text that was not in the brief.

Before a new-page brief reaches you, the ten results for its search are compared with the ten for the search an existing page already serves. Four or more shared addresses mean one intent, so the brief is refused and the action becomes expanding the page you have. Roof repair cost and how much to fix a roof are one page, and the module says so before you spend a day on the second.

The comparison is an intersection of two address lists, plain arithmetic with no model involved, and your own position plays no part in it. It counts as one external check in the allowance and is remembered for two weeks. Without paid result-page data the brief goes on, with a visible unchecked mark that says so.

Briefs that share a search stem form a cluster: a hub, its supporting pages and a publish set. The hub is the one page to draft now; the supporting briefs are shown as queued behind it until the hub has been read and indexed. The reading asks for one page at a time. A cluster is created only when its hub has a search with impressions behind it, which is the guard against pages nobody asked for.

An approved brief also gets a hold list: three to five of your existing pages that could honestly link to the new one, each with the paragraph and a phrase already in it. It is a list of candidates; the links come later. Once the new page has been read, the holds become candidates for the Internal Link Dealer, which accepts or refuses them under its usual rules and writes the verdicts back. Nothing is pasted in as a see-also sentence, and every anchor is a phrase already in the text.

Paste up to five competitor sitemap addresses. Only the address lists are fetched, and the pages themselves stay unread, and each address gets a search label inferred from its slug, marked as inferred. Opaque addresses that are only identifiers may get a label from a model, and that is marked too. A label says what a page seems to be about, and nothing about volume or position.

The result is three piles: page types they have and you do not, pages you have and they do not, and pages you both have where yours is losing impressions. Only the third pile stands on your own Search Console, so it is the only one that points at a page to work on, and it attaches to an opportunity you already have. The first pile is a backlog without a cap, outside the reading.

Competitor sitemaps show what others publish. This check shows which searches they win while you are absent from the results. It takes up to five competitors that share searches with you, lists the first-page searches they rank for and you do not, and keeps those with demand and a difficulty a smaller site can handle. Searches you already show for in Search Console are set aside, because the gap rule above already speaks about them.

Each remaining search is checked on the live result page. It stays only if that page holds at least two independent sites, with marketplaces and platforms left out, and one of them has few enough referring domains to be passed. That address is copied from the result page and shown next to the search. A search without one is dropped, and at most ten are kept. If nothing survives, the screen says at which filter the list ran dry.

You start it by hand on the Competitors screen, because every run spends checks from the site's allowance: at most 47, and a repeat within a month costs almost nothing. When you named the competitors by pasting their sitemaps, each search it keeps becomes a page to write and gets a brief on the next nightly pass; with competitors picked from search data, the list stays a report on this screen. A page to write stays outside this week's five. It has no forecast, because the site has no history for it yet.

Every other SEOLift screen reads Search Console. For a site without it, the Keywords (estimate) screen fetches the searches the domain ranks for from the DataForSEO index: position, page, monthly volume, difficulty and intent, the largest thousand first. Searches at positions 4 to 20 with at least 30 searches a month, and not your brand, are marked within reach and sorted by the clicks position 3 would add on a standard click curve. Pages are grouped by what they carry, and each fetch is compared with the one before: new, lost, up and down.

It is an estimate, and the screen says so above the numbers. The index knows searches with recorded volume, the position is its last check, and there is no CTR, because the index cannot know one. A fetch costs one check per thousand searches, a repeat within a week is free, and after the first it refreshes weekly. Connect Search Console and the estimate steps aside for the real numbers. Everything exports as CSV.

Your Search Console clusters already say which page serves which searches. The cluster map groups those pages into topics by the words they rank for: the page with the most impressions is the hub, the rest are its supporting pages. For every supporting page the map answers two questions from the latest reading: does it link to the hub, and does the hub link back. A link counts when the reading saw an internal, followed link in the page; a link that sits on every page is marked sitewide, because navigation is not a choice the topic made. It is built from the reading alone: nothing is fetched, bought or asked of a model.

The missing links are the list: which page should link to which, ordered by the demand behind the topic, exportable as CSV. With the Internal Link Dealer module the pairs go to its next run on their own: it finds the paragraph and the phrase on each page and judges the link under its usual rules, the pair like any other. The map is rebuilt after every nightly analysis and whenever you ask, and it says which reading its links were read from. And a recommendation that asks for internal links to a page of a topic names these pages as its first sources, read from the pages themselves.

Under the map, Room for one more page: for the largest topics, the searches that contain the hub's own subject, from the DataForSEO index, minus what you already show for, your brand, your excluded words and anything a page of yours already answers in other words. Each survivor is checked on the live result page against the hub's page and the pages it could be a rephrasing of; it stays only when Google shows different pages for it. What stays becomes a page to write with the hub as the page to link from. It runs when you ask and says what it would spend first; the seed is always the hub's own subject, and there is no field to type one.

Pages losing clicks against their own history of ninety days earlier already sit under Win back lost traffic. When the cause is content, with a canonical, a noindex or a slow response ruled out, the opportunity now names the searches that fell and the one section of the page that answers them. The draft is that section: update the stat, the price or the screenshot in it, then run the one-line IndexNow request from the draft. Fifteen minutes to a day of work, on one section.

With server logs uploaded, the verification after your change also says whether Googlebot and Bingbot fetched the page again, and when. Drops with a technical cause stay in the technical reading; content refresh is what is left.